ICPRAM 2013 Abstracts


Area 1 - Theory and Methods

Full Papers
Paper Nr: 23
Title:

Face Recognition using Modified Generalized Hough Transform and Gradient Distance Descriptor

Authors:

Marian Moise, Xue-Dong Yang and Richard Dosselman

Abstract: This research uses a modified version of the generalized Hough transform based on a new image descriptor, known as the gradient distance descriptor, to tackle the problem of face recognition. Thus, in addition to the position of the edges in a sketch of a face, this approach also takes into consideration the value of the corresponding descriptors. Individual descriptors are compared against one another using the matrix cosine similarity measure. This enables the technique to identify the region of a query face image that best matches a target face image in a database. The proposed technique does not require any training data and can be extended to general object recognition.

Paper Nr: 39
Title:

Improved Boosting Performance by Exclusion of Ambiguous Positive Examples

Authors:

Miroslav Kobetski and Josephine Sullivan

Abstract: In visual object class recognition it is difficult to densely sample the set of positive examples. Therefore, frequently there will be areas of the feature space that are sparsely populated, in which uncommon examples are hard to disambiguate from surrounding negatives without overfitting. Boosting in particular struggles to learn optimal decision boundaries in the presence of such hard and ambiguous examples. We propose a twopass dataset pruning method for identifying ambiguous examples and subjecting them to an exclusion function, in order to obtain more optimal decision boundaries for existing boosting algorithms. We also provide an experimental comparison of different boosting algorithms on the VOC2007 dataset, training them with and without our proposed extension. Using our exclusion extension improves the performance of all the tested boosting algorithms except TangentBoost, without adding any additional test-time cost. In our experiments LogitBoost performs best overall and is also significantly improved by our extension. Our results also suggest that outlier exclusion is complementary to positive jittering and hard negative mining.

Paper Nr: 54
Title:

Fourier Spectral of PalmCode as Descriptor for Palmprint Recognition

Authors:

Meiru Mu, Qiuqi Ruan, Luuk Spreeuwers and Raymond Veldhuis

Abstract: Study on automatic person recognition by palmprint is currently a hot topic. In this paper, we propose a novel palmprint recognition method by transforming the typical palmprint phase code feature into its Fourier frequency domain. The resulting real-valued Fourier spectral features are further processed by horizontal and vertical 2DPCA method, which proves highly efficient in terms of computational complexity, storage requirement and recognition accuracy. This paper also gives a contrast study on palm code and competitive code under the proposed feature extraction framework. Besides, experimental results on the Hongkong PolyU Palmprint database demonstrate that the proposed method outperforms many currently reported local Gabor pattern approaches for palmprint recognition.

Paper Nr: 63
Title:

Applications of Discriminative Dimensionality Reduction

Authors:

Barbara Hammer, Andrej Gisbrecht and Alexander Schulz

Abstract: Discriminative nonlinear dimensionality reduction aims at a visualization of a given set of data such that the information contained in the data points which is of particular relevance for a given class labeling is displayed. We link this task to an integration of the Fisher information, and we discuss its difference from supervised classification. We present two potential application areas: speed-up of unsupervised nonlinear visualization by integration of prior knowledge, and visualization of a given classifier such as an SVM in low dimensions.

Paper Nr: 70
Title:

An Adaptive Incremental Clustering Method based on the Growing Neural Gas Algorithm

Authors:

Mohamed-Rafik Bouguelia, Yolande Belaïd and Abdel Belaïd

Abstract: Usually, incremental algorithms for data streams clustering not only suffer from sensitive initialization parameters, but also incorrectly represent large classes by many cluster representatives, which leads to decrease the computational efficiency over time. We propose in this paper an incremental clustering algorithm based on ”growing neural gas” (GNG), which addresses this issue by using a parameter-free adaptive threshold to produce representatives and a distance-based probabilistic criterion to eventually condense them. Experiments show that the proposed algorithm is competitive with existing algorithms of the same family, while maintaining fewer representatives and being independent of sensitive parameters.

Paper Nr: 110
Title:

The Path Kernel

Authors:

Andrea Baisero, Florian T. Pokorny, Danica Kragic and Carl Henrik Ek

Abstract: Kernel methods have been used very successfully to classify data in various application domains. Traditionally, kernels have been constructed mainly for vectorial data defined on a specific vector space. Much less work has been addressing the development of kernel functions for non-vectorial data. In this paper, we present a new kernel for encoding sequential data. We present our results comparing the proposed kernel to the state of the art, showing a significant improvement in classification and a much improved robustness and interpretability.

Paper Nr: 116
Title:

Probabilistic Evidence Accumulation for Clustering Ensembles

Authors:

André Lourenço, Samuel Rota Bulò, Nicola Rebagliati, Ana Fred, Mário Figueiredo and Marcello Pelillo

Abstract: Ensemble clustering methods derive a consensus partition of a set of objects starting from the results of a collection of base clustering algorithms forming the ensemble. Each partition in the ensemble provides a set of pairwise observations of the co-occurrence of objects in a same cluster. The evidence accumulation clustering paradigm uses these co-occurrence statistics to derive a similarity matrix, referred to as co-association matrix, which is fed to a pairwise similarity clustering algorithm to obtain a final consensus clustering. The advantage of this solution is the avoidance of the label correspondence problem, which affects other ensemble clustering schemes. In this paper we derive a principled approach for the extraction of a consensus clustering from the observations encoded in the co-association matrix. We introduce a probabilistic model for the co-association matrix parameterized by the unknown assignments of objects to clusters, which are in turn estimated using a maximum likelihood approach. Additionally, we propose a novel algorithm to carry out the parameter estimation with convergence guarantees towards a local solution. Experiments on both synthetic and real benchmark data show the effectiveness of the proposed approach.

Paper Nr: 117
Title:

Relevance and Mutual Information-based Feature Discretization

Authors:

Artur Ferreira and Mario Figueiredo

Abstract: In many learning problems, feature discretization (FD) techniques yield compact data representations, which often lead to shorter training time and higher classification accuracy. In this paper, we propose two new FD techniques. The first method is based on the classical Linde-Buzo-Gray quantization algorithm, guided by a relevance criterion, and is able to work in unsupervised, supervised, or semi-supervised scenarios, depending on the adopted measure of relevance. The second method is a supervised technique based on the maximization of the mutual information between each discrete feature and the class label. For both methods, our experiments on standard benchmark datasets show their ability to scale up to high-dimensional data, attaining in many cases better accuracy than other FD approaches, while using fewer discretization intervals.

Paper Nr: 118
Title:

Multiclass Diffuse Interface Models for Semi-supervised Learning on Graphs

Authors:

Cristina Garcia-Cardona, Arjuna Flenner and Allon G. Percus

Abstract: We present a graph-based variational algorithm for multiclass classification of high-dimensional data, motivated by total variation techniques. The energy functional is based on a diffuse interface model with a periodic potential. We augment the model by introducing an alternative measure of smoothness that preserves symmetry among the class labels. Through this modification of the standard Laplacian, we construct an efficient multiclass method that allows for sharp transitions between classes. The experimental results demonstrate that our approach is competitive with the state of the art among other graph-based algorithms.

Paper Nr: 122
Title:

Discriminative Sequence Back-constrained GP-LVM for MOCAP based Action Recognition

Authors:

Valsamis Ntouskos, Panagiotis Papadakis and Fiora Pirri

Abstract: In this paper we address the problem of human action recognition within Motion Capture sequences. We introduce a method based on Gaussian Process Latent Variable Models and Alignment Kernels. We build a new discriminative latent variable model with back-constraints induced by the similarity of the original sequences. We compare the proposed method with a standard sequence classification method based on Dynamic Time Warping and with the recently introduced V-GPDS model which is able to model highly dimensional dynamical systems. The proposed methodology exhibits high performance even for datasets that have not been manually preprocessed while it further allows fast inference by exploiting the back constraints.

Paper Nr: 144
Title:

A 3D Segmentation Algorithm for Ellipsoidal Shapes - Application to Nuclei Extraction

Authors:

Emmanuel Soubies, Pierre Weiss and Xavier Descombes

Abstract: We propose some improvements of the Multiple Birth and Cut algorithm (MBC) in order to extract nuclei in 2D and 3D images. This algorithm based on marked point processes was proposed recently in (Gamal Eldin et al., 2012). We introduce a new contrast invariant energy that is robust to degradations encountered in fluorescence microscopy (e.g. local radiometry attenuations). Another contribution of this paper is a fast algorithm to determine whether two ellipses (2D) or ellipsoids (3D) intersect. Finally, we propose a new heuristic that strongly improves the convergence rates. The algorithm alternates between two birth steps. The first one consists in generating objects uniformly at random and the second one consists in perturbing the current configuration locally. Performance of this modified birth step is evaluated and examples on various image types show the wide applicability of the method in the field of bio-imaging.

Short Papers
Paper Nr: 3
Title:

Bayesian Regularized Committee of Extreme Learning Machine

Authors:

José M. Martínez-Martínez, Pablo Escandell-Montero, Emilio Soria-Olivas, Joan Vila-Francés and Rafael Magdalena-Benedito

Abstract: Extreme Learning Machine (ELM) is an efficient learning algorithm for Single-Hidden Layer Feedforward Networks (SLFNs). Its main advantage is its computational speed due to a random initialization of the parameters of the hidden layer, and the subsequent use of Moore-Penrose’s generalized inverse in order to compute the weights of the output layer. The main inconvenient of this technique is that as some parameters are randomly assigned and remain unchanged during the training process, they can be non-optimum and the network performance may be degraded. This paper aims to reduce this problem using ELM committees. The way to combine them is to use a Bayesian linear regression due to its advantages over other approaches. Simulations on different data sets have demonstrated that this algorithm generally outperforms the original ELM algorithm.

Paper Nr: 6
Title:

Latent Ambiguity in Latent Semantic Analysis?

Authors:

Martin Emms and Alfredo Maldonado-Guerra

Abstract: Latent Semantic Analyis (LSA) consists in the use of SVD-based dimensionality-reduction to reduce the high dimensionality of vector representations of documents, where the dimensions of the vectors correspond simply to word counts in the documents. We show that that there are two contending, inequivalent, formulations of LSA. The distinction between the two is not generally noted and while some work adheres to one formulation, other work adheres to the other formulation. We show that on both a tiny contrived data-set and also on a more substantial word-sense discovery data-set that the empirical outcomes achieved with LSA vary according to which formulation is chosen.

Paper Nr: 41
Title:

Accelerated Nonlinear Gaussianization for Feature Extraction

Authors:

Alexandru Paul Condurache and Alfred Mertins

Abstract: In a multi-class classification setup, the Gaussianization represents a nonlinear feature extraction transform with the purpose of achieving Gaussian class-conditional densities in the transformed space. The computational complexity of such a transformation increases with the dimension of the processed feature space in such a way that only relatively small dimensions can be processed. In this contribution we describe how to reduce the computational burden with the help of an adaptive grid. Thus, the Gaussianization transform is able to also handle feature spaces of higher dimensionality, improving upon its practical usability. On both artificially generated and real-application data, we demonstrate a decrease in computation complexity in comparison to the standard Gaussianization, while maintaining the effectiveness.

Paper Nr: 46
Title:

Desirability Function Approach on the Optimization of Multiple Bernoulli-distributed Response

Authors:

Frederick Kin Hing Phoa and Hsiu-Wen Chen

Abstract: The multiple response optimization (MRO) problem is commonly found in industry and many other scientific areas. During the optimization stage, the desirability function method, first proposed by Harrington (1965), has been widely used for optimizing multiple responses simultaneously. However, the formulation of traditional desirability functions breaks down when the responses are Bernoulli-distributed. This paper proposes a simple solution to avoid this breakdown. Instead of the original binary responses, their probabilities of defined outcomes are considered in the logistic regression models and they are transformed into the desirability functions. An example is used for demonstration.

Paper Nr: 48
Title:

On using Additional Unlabeled Data for Improving Dissimilarity-Based Classifications

Authors:

Sang-Woon Kim

Abstract: This paper reports an experimental result obtained with additionally using unlabeled data together with labeled ones to improve the classification accuracy of dissimilarity-based methods, namely, dissimilarity-based classifications (DBC) (Pe¸kalska, E. and Duin, R. P .W., 2005). In DBC, classifiers among classes are not based on the feature measurements of individual objects, but rather on a suitable dissimilarity measure among the objects. In image classification tasks, on the other hand, one of the intractable problems is the lack of information caused by the insufficient number of data. To address this problem in DBC, in this paper we study a new way of measuring the dissimilarity distance between two object images by using the well-known one-shot similarity metric (OSS) (Wolf, L. et al., 2009). In DBC using OSS, the dissimilarity distance is measured based on unlabeled (background) data that do not belong to the classes being learned, and consequently, do not require labeling. From this point of view, the classification is done in a semi-supervised learning (SSL) framework. Our experimental results, obtained with well-known benchmarks, demonstrate that when the cardinalities of the unlabeled data set and the prototype set have been appropriately chosen using additional unlabeled data for the OSS metric in SSL, DBC can be improved in terms of classification accuracies.

Paper Nr: 51
Title:

Clupea Harengus: Intraspecies Distinction using Curvature Scale Space and Shapelets - Classification of North-sea and Thames Herring using Boundary Contour of Sagittal Otoliths

Authors:

James Mapp, Mark Fisher, Anthony Bagnall, Jason Lines, Sally Warne and Joe Scutt Phillips

Abstract: We present a study comparing Curvature Scale Space (CSS) representation with Shapelet transformed data with a view to discriminating between sagittal otoliths of North-Sea and Thames Herring using otolith boundary and boundary metrics. CSS transformed boundaries combined with measures of their circularity, eccentricity and aspect-ratio are used to classify using nearest-neighbour selections with distance being computed using CSS matching methods. Shapelet transformed data are classified using a number of techniques (Nearest-Neighbour, Naive-Bayes, C4.5, Support Vector Machines, Random and Rotation Forest) and compared to CSS classification results. Both methods use Leave One Out Cross Validation (LOOCV). We describe the method of encoding and the matching algorithm used during CSS classification and give an overview of Shapelet transforms and the classifiers that are used on the data. It is shown that whilst CSS forms part of the MPEG-7 standard and performs better than random selection, it can be significantly out-performed by recent additions to machine-learning methods in this application. Shapelets also show that with regard to intra-species distinction, the discriminatory features of otolith boundaries may lie not in the major inflection points, but the boundary points between them.

Paper Nr: 61
Title:

Declarative Gesture Spotting using Inferred and Refined Control Points

Authors:

Lode Hoste, Brecht De Rooms and Beat Signer

Abstract: We propose a novel gesture spotting approach that offers a comprehensible representation of automatically inferred spatiotemporal constraints. These constraints can be defined between a number of characteristic control points which are automatically inferred from a single gesture sample. In contrast to existing solutions which are limited in time, our gesture spotting approach offers automated reasoning over a complete motion trajectory. Last but not least, we offer gesture developers full control over the gesture spotting task and enable them to refine the spotting process without major programming efforts.

Paper Nr: 62
Title:

Inverse of Lorentzian Mixture for Simultaneous Training of Prototypes and Weights

Authors:

Atsushi Sato and Masato Ishii

Abstract: This paper presents a novel distance-based classifier based on the multiplicative inverse of Lorentzian mixture, which can be regarded as a natural extension of the conventional nearest neighbor rule. We show that prototypes and weights can be trained simultaneously by General Loss Minimization, which is a generalized version of supervised learning framework used in Generalized Learning Vector Quantization. Experimental results for UCI machine learning repository reveal that the proposed method achieves almost the same as or higher classification accuracy than Support Vector Machine with a much fewer prototypes than support vectors.

Paper Nr: 66
Title:

A Statistic Criterion for Reducing Indeterminacy in Linear Causal Modeling

Authors:

Gianluca Bontempi

Abstract: Inferring causal relationships from observational data is still an open challenge in machine learning. State-of-the-art approaches often rely on constraint-based algorithms which detect v-structures in triplets of nodes in order to orient arcs. These algorithms are destined to fail when confronted with completely connected triplets. This paper proposes a criterion to deal with arc orientation also in presence of completely linearly connected triplets. This criterion is then used in a Relevance-Causal (RC) algorithm, which combines the original causal criterion with a relevance measure, to infer causal dependencies from observational data. A set of simulated experiments on the inference of the causal structure of linear networks shows the effectiveness of the proposed approach.

Paper Nr: 68
Title:

Robust 3-D Object Skeletonisation for the Similarity Measure

Authors:

Christian Feinen, David Barnowsky, Dietrich Paulus and Marcin Grzegorzek

Abstract: In this paper we introduce our approach for similarity measure of 3-D objects based on an existing curve skeletonization technique. This skeletonization algorithm for 3-D objects delivers skeletons thicker than 1 voxel. This makes an efficient distance or similarity measure impossible. To overcome this drawback, we use a significantly extended skeletonization algorithm (by Reniers) and a modified Dijkstra approach. In addition to that, we propose features that are directly extracted from the resulting skeletal structures. To evaluate our system, we created a ground truth of 3-D objects and their similarities estimated by humans. The automatic similarity results achieved by our system were evaluated against this ground truth in terms of precision and recall in an object retrieval setup.

Paper Nr: 90
Title:

Iterative Possibility Distributions Refining in Pixel-based Images Classification Framework

Authors:

B. Alsahwa, S. Almouahed, D. Guériot and B. Solaiman

Abstract: In this study, an incremental and iterative approach for possibility distributions estimation in pixel-based images classification context is proposed. This approach is based on the use of possibilistic reasoning in order to enrich a set of samples serving for the initial estimation of possibility distributions. The use of possibilistic concepts enables an important flexibility for the integration of a context-based additional semantic knowledge source formed by pixels belonging with high certainty to different semantic classes (called possibilistic seeds), into the available knowledge encoded by possibility distributions. Once possibilistic seeds are extracted, possibility distributions are incrementally updated and refined. Synthetic images composed of two thematic classes are generated in order to evaluate the performances of the proposed approach. Initial possibility distributions are, first, obtained using a priori knowledge given in the form of learning areas delimitated by an expert. These areas serve for the estimation of the probability distributions of different thematic classes. The resulting probability density functions are then transformed into possibility distributions using Dubois-Prade’s probability-possibility transformation. The possibilistic seeds extraction process is conducted through the application of a possibilistic contextual rule using the confidence index used as an uncertainty measure.

Paper Nr: 95
Title:

Passive-aggressive Online Learning for Relevance Feedback in Content based Image Retrieval

Authors:

Luca Piras, Giorgio Giacinto and Roberto Paredes

Abstract: The increasing availability of large archives of digital images has pushed the need for effective image retrieval systems. Relevance Feedback (RF) techniques, where the user is involved in an iterative process to refine the search, have been recently formulated in terms of classification paradigms in low-level feature spaces. Two main issues arises in this formulation, namely the small size of the training set, and the unbalance between the class of relevant images and all other non-relevant images. To address these issues, in this paper we propose to formulate the RF paradigm in terms of Passive-Aggressive on-line learning approaches. These approaches are particularly suited to be implemented in RF because of their iterative nature, which allows further improvements in the image search process. The reported results show that the performances attained by the proposed algorithm are comparable, and in many cases higher, than those attained by other RF approaches.

Paper Nr: 107
Title:

Qualitative Vocabulary based Descriptor

Authors:

Heydar Maboudi Afkham, Carl Henrik Ek and Stefan Carlsson

Abstract: Creating a single feature descriptors from a collection of feature responses is an often occurring task. As such the bag-of-words descriptors have been very successful and applied to data from a large range of different domains. Central to this approach is making an association of features to words. In this paper we present a new and novel approach to feature to word association problem. The proposed method creates a more robust representation when data is noisy and requires less words compared to the traditional methods while retaining similar performance. We experimentally evaluate the method on a challenging image classification data-set and show significant improvement to the state of the art.

Paper Nr: 119
Title:

Classification of Dissimilarity Data via Sparse Representation

Authors:

Ilias Theodorakopoulos, George Economou and Spiros Fotopoulos

Abstract: Pairwise dissimilarity representations are common practice in several applications of computer vision, since they constitute a powerful alternative to the traditional vectorial representations. During the previous decades many techniques aiming to quantify the dissimilarity between two objects were developed. The problem is that most of these measures tend to produce non-Euclidean and/or non-metric dissimilarity data, although they seem to perform quite well on a variety of tasks. Recently it has been shown that non-Euclidean properties of dissimilarity data may contain useful and discriminative information. In this context, classical embedding of dissimilarity data into a vector space, can imply information loss. An alternative option is the representation into the dissimilarity space, where each object is represented by it’s dissimilarity to a set of prototypes. Such a space has the mathematical properties which allow the incorporation of more advanced classifiers, beyond the Nearest Neighbour and the k-NN which are usually the case. In the current work we aim to combine the flexibility of dissimilarity representations with the discriminative ability of the well-established sparse representation-based classification scheme (Wright, 2010), in order to enhance the classification performance on dissimilarity data. The proposed DS-SRC framework has been evaluated on three datasets, derived from different computer vision tasks. The results demonstrate the ability of DS-SRC to improve the classification accuracy, regardless of the special characteristics of each dataset.

Paper Nr: 125
Title:

A Tensor-based Clustering Approach for Multiple Document Classifications

Authors:

Salvatore Romeo, Andrea Tagarelli, Francesco Gullo and Sergio Greco

Abstract: We propose a novel approach to the problem of document clustering when multiple organizations are provided for the documents in input. Besides considering the information on the text-based content of the documents, our approach exploits frequent associations of the documents in the groups across the existing classifications, in order to capture how documents tend to be grouped together orthogonally to different views. A third-order tensor for the document collection is built over both the space of terms and the space of the discovered frequent document-associations, and then it is decomposed to finally establish a unique encompassing clustering of documents. Preliminary experiments conducted on a document clustering benchmark have shown the potential of the approach to capture the multi-view structure of existing organizations for a given collection of documents.

Paper Nr: 126
Title:

Evolving Classifier Ensembles using Dynamic Multi-objective Swarm Intelligence

Authors:

Jean-François Connolly, Eric Granger and Robert Sabourin

Abstract: Classification systems are often designed using a limited amount of data from complex and changing pattern recognition environments. In applications where new reference samples become available over time, adaptive multi-classifier systems (AMCSs) are desirable for updating class models. In this paper, an incremental learning strategy based on an aggregated dynamical niching particle swarm optimization (ADNPSO) algorithm is proposed to efficiently evolve heterogeneous classifier ensembles in response to new reference data. This strategy is applied to an AMCS where all parameters of a pool of fuzzy ARTMAP (FAM) neural network classifiers, each one corresponding to a PSO particle, are co-optimized such that both error rate and network size are minimized. To sustain a high level of accuracy while minimizing the computational complexity, the AMCS integrates information from multiple diverse classifiers, where learning is guided by the ADNPSO algorithm that optimizes networks according both these objectives. Moreover, FAM networks are evolved to maintain (1) genotype diversity of solutions around local optima in the optimization search space, and (2) phenotype diversity in the objective space. Using local Pareto optimality, networks are then stored in an archive to create a pool of base classifiers among which cost-effective ensembles are selected on the basis of accuracy, and both genotype and phenotype diversity. Performance of the ADNPSO strategy is compared against AMCSs where learning of FAM networks is guided through mono- and multi-objective optimization, and assessed under different incremental learning scenarios, where new data is extracted from real-world video streams for face recognition. Simulation results indicate that the proposed strategy provides a level of accuracy that is comparable to that of using mono-objective optimization, yet requires only a fraction of its resources.

Paper Nr: 142
Title:

A Novel Regression Method for Software Defect Prediction with Kernel Methods

Authors:

Ahmet Okutan and Olcay Taner Yıldız

Abstract: In this paper, we propose a novel method based on SVM to predict the number of defects in the files or classes of a software system. To model the relationship between source code similarity and defectiveness, we use SVM with a precomputed kernel matrix. Each value in the kernel matrix shows how much similarity exists between the files or classes of the software system tested. The experiments on 10 Promise datasets indicate that SVM with a precomputed kernel performs as good as the SVM with the usual linear or RBF kernels in terms of the root mean square error (RMSE). The method proposed is also comparable with other regression methods like linear regression and IBK. The results of this study suggest that source code similarity is a good means of predicting the number of defects in software modules. Based on the results of our analysis, the developers can focus on more defective modules rather than on less or non defective ones during testing activities.

Paper Nr: 145
Title:

Spatially Varying Blur Recovery - Diagonal Approximations in the Wavelet Domain

Authors:

Paul Escande, Pierre Weiss and Francois Malgouyres

Abstract: Restoration of images degraded by spatially varying blurs is an issue of increasing importance. Many new optical systems allow to know the system point spread function at some random locations, by using microscopic luminescent structures. Given a set of impulse responses, we propose a fast and efficient algorithm to reconstruct the blurring operator in the whole image domain. Our method consists in finding an approximation of the integral operator by operators diagonal in the wavelet domain. Interestingly, this method complexity scales linearly with the image size. It is thus applicable to large 3D problems. We show that this approach might outperform previously proposed strategies such as linear interpolations (Nagy and O’Leary, 1998) or separable approximations (Zhang et al., 2007). We provide various theoretical and numerical results in order to justify the proposed methods.

Paper Nr: 149
Title:

Evaluating Learning Algorithms for Stochastic Finite Automata - Comparative Empirical Analyses on Learning Models for Technical Systems

Authors:

Asmir Vodenčarević, Alexander Maier and Oliver Niggemann

Abstract: Finite automata are used to model a large variety of technical systems and form the basis of important tasks such as model-based development, early simulations and model-based diagnosis. However, such models are today still mostly derived manually, in an expensive and time-consuming manner. Therefore in the past twenty years, several successful algorithms have been developed for learning various types of finite automata. These algorithms use measurements of the technical systems to automatically derive the underlying automata models. However, today users face a serious problem when looking for such model learning algorithm: Which algorithm to choose for which problem and which technical system? This papers closes this gap by comparative empirical analyses of the most popular algorithms (i) using two real-world production facilities and (ii) using artificial datasets to analyze the algorithms’ convergence and scalability. Finally, based on these results, several observations for choosing an appropriate automaton learning algorithm for a specific problem are given.

Paper Nr: 151
Title:

Image Pyramids as a New Approach for the Determination of Fractal Dimensions

Authors:

Michael Mayrhofer-Reinhartshuber, Philipp Kainz and Helmut Ahammer

Abstract: The consideration of different scales and the application of fractal methods on digital images is of high importance if real world objects are investigated. In this context the fractal dimension is an important parameter to characterize structures and patterns. An accurate understanding of them is obligatory if significant and comparable results should be obtained. Recently a new method using an image pyramid approach was compared to the very popular Box Counting Method. The intriguing results showed that a trustable value for the fractal dimension could be obtained in much faster computational times compared to traditional Box Counting algorithms. In addition to these results of this new approach, which is only applicable to binary (black/white) images, we present developments toward the application to grey value/color images. Especially the determination of the grey value surface and the interpolation used to downscale the images seem to have major influence on the results achieved.

Paper Nr: 167
Title:

HMM based Classifier for the Recognition of Roots of a Large Canonical Arabic Vocabulary

Authors:

Imen Ben Cheikh and Zeineb Zouaoui

Abstract: The complexity of the recognition process is strongly related to language, the type of writing and the vocabulary size. Our work represents a contribution to a system of recognition of large canonical Arabic vocabulary of decomposable words derived from tri-consonantal roots. This system is based on a collaboration of three morphological classifiers specialized in the recognition of roots, schemes and conjugations. Our work deals with the first classifier. It is about proposing a root classifier based on 101 Hidden Markov Models, used to classify 101 tri-consonantal roots. The models have the same architecture endowed with Arabic linguistic knowledge. The proposed system deals, up to now, with a vocabulary of 5757 words. It has been learned then tested using a total of more than 17000 samples of printed words. Obtained results are satisfying and the top2 recognition rate reached 96%.

Posters
Paper Nr: 14
Title:

A Flexible Particle Swarm Optimization based on Global Best and Global Worst Information

Authors:

Emre Çomak

Abstract: A reverse direction supported particle swarm optimization (RDS-PSO) method was proposed in this paper. The main idea to create such a method relies that on benefiting from global worst particle in reverse direction. It offers avoiding from local optimal solutions and providing diversity thanks to its flexible velocity update equation. Various experimental studies have been done in order to evaluate the effect of variable inertia weight parameter on RDS-PSO by using of Rosenbrock, Rastrigin, Griewangk and Ackley test functions. Experimental results showed that RDS-PSO, executed with increasing inertia weight, offered relatively better performance than RDS-PSO with decreasing one. RDS-PSO executed with increasing inertia weight produced remarkable improvements except on Rastrigin function when it is compared with original PSO.

Paper Nr: 17
Title:

On the Strategy to Follow for Skeleton Pruning - Sequential, Parallel and Hybrid Approaches

Authors:

Maria Frucci, Gabriella Sanniti di Baja, Carlo Arcelli and Luigi P. Cordella

Abstract: Pruning is an important step in a skeletonization process and a number of pruning criteria have been suggested in the literature. However, the modality to be followed when checking the pruning criterion is not generally described in detail. In our opinion, two main pruning modalities can be envisaged and in this paper we discuss their impact on the performance of pruning. Moreover, we introduce a third modality, which we regard as able to provide a more satisfactory pruning performance.

Paper Nr: 29
Title:

3D Model Retrieval using Density-based Silhouette Descriptor

Authors:

Qi Tang and Xin Yang

Abstract: In this paper we present a new content-based retrieval descriptor, density-based silhouette descriptor (DBS). It characterizes a 3D object with multivariate probability functions of its 2D silhouette features. The new descriptor is computationally efficient and induces a permutation property that guarantees invariance at the matching stage. Also, it is insensitive to small shape perturbations and mesh resolution. The retrieval performance on several 3D databases shows that the DBS provides state-of-art discrimination over a broad and heterogeneous set of shape categories.

Paper Nr: 91
Title:

Possibilistic Similarity based Image Classification

Authors:

B. Alsahwa, S. Almouahed, D. Guériot and B. Solaiman

Abstract: In this study, an approach for image classification based on possibilistic similarity is proposed. This approach, due to the use of possibilistic concepts, enables an important flexibility to integrate both contextual information and a priori knowledge. Possibility distributions are, first, obtained using a priori knowledge given in the form of learning areas delimitated by an expert. These areas serve for the estimation of the probability density functions of different thematic classes. The resulting probability density functions are then transformed into possibility distributions using Dubois-Prade’s probability-possibility transformation. Several measures of similarity between classes were tested in order to improve the discrimination between classes. The classification is then performed based on the principle of possibilistic similarity. Synthetic and real images are used in order to evaluate the performances of the proposed model.

Paper Nr: 96
Title:

The Area under the ROC Curve as a Criterion for Clustering Evaluation

Authors:

Helena Aidos, Robert P. W. Duin and Ana Fred

Abstract: In the literature, there are several criteria for validation of a clustering partition. Those criteria can be external or internal, depending on whether we use prior information about the true class labels or only the data itself. All these criteria assume a fixed number of clusters k and measure the performance of a clustering algorithm for that k. Instead, we propose a measure that provides the robustness of an algorithm for several values of k, which constructs a ROC curve and measures the area under that curve. We present ROC curves of a few clustering algorithms for several synthetic and real-world datasets and show which clustering algorithms are less sensitive to the choice of the number of clusters, k. We also show that this measure can be used as a validation criterion in a semi-supervised context, and empirical evidence shows that we do not need always all the objects labeled to validate the clustering partition.

Paper Nr: 97
Title:

Arabic Character Recognition based on Statistical Features - A Comparative Study

Authors:

Mariem Gargouri Kchaou, Slim Kanoun, Fouad Slimane and Souhir Bouaziz Affes

Abstract: This paper presents a comparative study for Arabic optical character recognition techniques according to statistic approach. So, the current work consists in experimenting character image characterization and matching to show the most robust and reliable techniques. For features extraction phase, we test invariant moments, affine moment invariants, Tsirikolias–Mertzios moments, Zernike moments, Fourier-Mellin transform and Fourier descriptors. And for the classification phase, we use k-Nearest Neighbors and Support Vector Machine. Our data collection encloses 3 datasets. The first contains 2320 multi-font and multi-scale printed samples. The second contains 9280 multi-font, multi-scale and multi-oriented printed samples. And, the third contains 2900 handwritten samples which are extracted from the IFN/ENIT data. The aim was to cover a wide spectrum of Arabic characters complexity. The best performance rates found for each dataset are 99.91%, 99.26% and 66.68% respectively.

Paper Nr: 99
Title:

A Novel Clustering-Based Ensemble Classification Model for Block Learning

Authors:

Mohammad Raihanul Islam, Mustafizur Rahman, Asif Salekin, Shihab Hasan Chowdhury and Samiul Alam Anik

Abstract: In this paper, we have considered a real life scenario where data is available in blocks over the period of time. We have developed a dynamic cluster based ensemble of classifiers for the problem. We have applied clustering algorithm on the block of data available at that time and have trained a neural network for each of the clusters. The performance of the network is tested against the next available block of data and based on that performance the parameters of the clustering algorithm is changed at runtime. In our approach increasing the number of clusters is considered as changing of the parameter settings. The misclassified instances of the test data are also joined with the training data to refine the knowledge of the classifiers. The proposed system is capable of identifying the decision boundary of different classes based on the current block of data more precisely. An extensive experiments has been performed to evaluate this dynamic system and to compute the optimal parameters of the proposed procedure.

Paper Nr: 100
Title:

Multistage Naive Bayes Classifier with Reject Option for Multiresolution Signal Representation

Authors:

Urszula Libal

Abstract: In the article, two approaches to pattern recognition of signals are compared: a direct and a multistage. It is assumed that there are two generic patterns of signals, i.e. a two-class problem is considered. The direct method classifies signal in one step. The multistage method uses a multiresolution representation of signal in wavelet bases, starting from a coarse resolution at the first stage to a more detailed resolutions at the next stages. After a signal is assigned to a class, the posterior probability for this class is counted and compared with a fixed level. If the probability is higher than this level, the algorithm stops. Otherwise the signal is rejected and on the next stage the classification procedure is repeated for a higher resolution of signal. The posterior probability is calculated again. The algorithm stops when the probability is higher than a fixed level and a signal is finally assigned to a class. The wavelet filtration of signal is used for feature selection and acts as a magnifier. If the posterior probability of recognition is low on some stage, the number of features on the next stage is increased by taking a better resolution. The experiments are performed for three local decision rules: naive Bayes, linear and quadratic discriminant analysis.

Paper Nr: 105
Title:

Predicting Classifier Combinations

Authors:

Matthias Reif, Annika Leveringhaus, Faisal Shafait and Andreas Dengel

Abstract: Combining classifiers is a common technique in order to improve the performance and robustness of classification systems. However, the set of classifiers that should be combined is not obvious and either expert knowledge or a time consuming evaluation phase is required in order to achieve high accuracy values. In this paper, we present an approach of automatically selecting the set of base classifiers for combination. The method uses experience about previous classifier combinations and characteristics of datasets in order to create a prediction model. We evaluate the method on over 80 datasets. The results show that the presented method is able to reasonably predict a suitable set of base classifiers for most of the datasets.

Paper Nr: 106
Title:

A Novel Adaptive Fuzzy Model for Image Retrieval

Authors:

Payam Pourashraf, Mohsen Ebrahimi Moghaddam and Saeed Bagheri Shouraki

Abstract: In many areas of commerce, medicine, entertainment, education, weather forecasting the need for efficient image retrieval system has grown dramatically. Therefore, many researches have been done in this scope; however, researchers try to improve the precision and performance of such system. In this paper, we present an image retrieval method, which uses color and texture based approaches for feature extraction, fuzzy adaptive model and fuzzy integral. The system extracts color and texture features from an image and enhancing the retrieval by providing a unique adaptive fuzzy system that use fuzzy membership functions to find the region of interest in an image. The proposed method aggregates the features by assigning fuzzy measures and combines them with the help of fuzzy integral. Experimental results showed that proposed method has some advantages and better results versus related ones in most of the time.

Paper Nr: 109
Title:

Nonlinearity Reduction of Manifolds using Gaussian Blur for Handshape Recognition based on Multi-Dimensional Grids

Authors:

Mohamed Farouk, Alistair Sutherland and Amin Shokry

Abstract: This paper presents a hand-shape recognition algorithm based on using multi-dimensional grids (MDGs) to divide the feature space of a set of hand images. Principal Component Analysis (PCA) is used as a feature extraction and dimensionality reduction method to generate eigenspaces from example images. Images are blurred by convolving with a Gaussian kernel as a low pass filter. Image blurring is used to reduce the non-linearity in the manifolds within the eigenspaces where MDG structure can be used to divide the spaces linearly. The algorithm is invariant to linear transformations like rotation and translation. Computer generated images for different hand-shapes in Irish Sign Language are used in testing. Experimental results show accuracy and performance of the proposed algorithm in terms of blurring level and MDG size.

Paper Nr: 112
Title:

Exploiting Correlation-based Metrics to Assess Encoding Techniques

Authors:

Giuliano Armano and Emanuele Tamponi

Abstract: The performance of a classification system depends on various aspects, including encoding techniques. In fact, encoding techniques play a primary role in the process of tuning a classifier/predictor, as choosing the most appropriate encoder may greatly affect its performance. As of now, evaluating the impact of an encoding technique on a classification system typically requires to train the system and test it by means of a performance metric deemed relevant (e.g., precision, recall, and Matthews correlation coefficients). For this reason, assessing a single encoding technique is a time consuming activity, which introduces some additional degrees of freedom (e.g., parameters of the training algorithm) that may be uncorrelated with the encoding technique to be assessed. In this paper, we propose a family of methods to measure the performance of encoding techniques used in classification tasks, based on the correlation between encoded input data and the corrisponding output. The proposed approach provides correlation-based metrics, devised with the primary goal of focusing on the encoding technique, leading other unrelated aspects apart. Notably, the proposed technique allows to save computational time to a great extent, as it needs only a tiny fraction of the time required by standard methods.

Paper Nr: 114
Title:

Graph-based Shape Representation for Object Retrieval

Authors:

Ali Amanpourgharaei, Christian Feinen and Marcin Grzegorzek

Abstract: Shape analysis has been an area of interest and research in image processing for a long time. Developing a discriminant shape representation and description method is a concern in many applications like image retrieval systems. This paper presents a new shape representation model which is based on graphs. We also present developed similarity measure technique to find correspondences between shapes. In our approach, features extracted from boundary of the shape are used to build up a graph. By means of a novel solution for attributed graph matching a new method for shape similarity measure is built up.

Paper Nr: 130
Title:

A System of Cursive Basic-word Recognition using HMMS for Bank-check Processing

Authors:

G. Facchini

Abstract: This paper presents a new system of cursive basic-word recognition for the legal amount of bank-check based on Hidden Markov Models and the sliding window technique. From off-line image analysis, the tracing of the handwritten words highlights the singularities that occurred; this information can be used for the reduction of the lexicon. This is a statistical recognition segmentation-free method for the basic words. The approach uses specific singularity markers to support the recognition phase. Two strategies for sliding window step are considered, the regular step and the progressive step, to capture many types of features in order to transform them into useful information in the modelling of the HMMs. Several experiments have been performed using the SIDB benchmark database of our laboratory. Experimental results show the improvements obtained for basic word lexicon recognition.

Paper Nr: 156
Title:

Smoothing Parameters Selection for Dimensionality Reduction Method based on Probabilistic Distance - Application to Handwritten Recognition

Authors:

Faycel El Ayeb and Faouzi Ghorbel

Abstract: Here, we intend to give a rule for the choice of the smoothing parameter of the orthogonal estimate of Patrick-Fisher distance in the sense of the Mean Integrate Square Error. The orthogonal series density estimate precision depends strongly on the choice of such parameter which corresponds to the number of terms in the series expansion used. By using series of random simulations, we illustrate the better performance of its dimensionality reduction in the mean of the misclassification rate. We show also its better behavior for real data. Different invariant shape descriptors describing handwritten digits are extracted from a large database. It serves to compare the proposed adjusted Patrick-Fisher distance estimator with a conventional feature selection method in the mean of the probability error of classification.

Area 2 - Applications

Full Papers
Paper Nr: 24
Title:

Efficient Bag of Scenes Analysis for Image Categorization

Authors:

Sébastien Paris, Xanadu Halkias and Hervé Glotin

Abstract: In this paper, we address the general problem of image/object categorization with a novel approach referred to as Bag-of-Scenes (BoS).Our approach is efficient for low semantic applications such as texture classification as well as for higher semantic tasks such as natural scenes recognition or fine-grained visual categorization (FGVC). It is based on the widely used combination of i) Sparse coding (Sc), ii) Max-pooling and iii) Spatial Pyramid Matching (SPM) techniques applied to histograms of multi-scale Local Binary/Ternary Patterns (LBP/LTP) and its improved variants. This approach can be considered as a two-layer hierarchical architecture: the first layer encodes the local spatial patch structure via histograms of LBP/LTP while the second encodes the relationships between pre-analyzed LBP/LTP-scenes/objects. Our method outperforms SIFT-based approaches using Sc techniques and can be trained efficiently with a simple linear SVM.

Paper Nr: 53
Title:

Unsupervised and Transfer Learning under Uncertainty - From Object Detections to Scene Categorization

Authors:

Grégoire Mesnil, Salah Rifai, Antoine Bordes , Xavier Glorot, Yoshua Bengio and Pascal Vincent

Abstract: Classifying scenes (e.g. into “street”, “home” or “leisure”) is an important but complicated task nowadays, because images come with variability, ambiguity, and a wide range of illumination or scale conditions. Standard approaches build an intermediate representation of the global image and learn classifiers on it. Recently, it has been proposed to depict an image as an aggregation of its contained objects:the representation on which classifiers are trained is composed of many heterogeneous feature vectors derived from various object detectors. In this paper, we propose to study different approaches to efficiently combine the data extracted by these detectors. We use the features provided by Object-Bank (Li-Jia Li and Fei-Fei, 2010a) (177 different object detectors producing 252 attributes each), and show on several benchmarks for scene categorization that careful combinations, taking into account the structure of the data, allows to greatly improve over original results (from +5% to +11%) while drastically reducing the dimensionality of the representation by 97% (from 44;604 to 1; 000).

Paper Nr: 67
Title:

Cluster Detection and Field-of-View Quality Rating - Applied to Automated Pap-smear Analysis

Authors:

Marine Astruc, Patrik Malm, Rajesh Kumar and Ewert Bengtsson

Abstract: Automated cervical cancer screening systems require high resolution analysis of a large number of epithelial cells, involving complex algorithms, mainly analysing the shape and texture of cell nuclei. This can be a very time consuming process. An initial selection of relevant fields-of-view in low resolution images could limit the number of fields to be further analysed at a high resolution. In particular, the detection of cell clusters is of interest for nuclei segmentation improvement, and for diagnostic purpose, malignant and endometrial cells being more prone to stick together in clusters than other cells. In this paper, we propose methods aiming at evaluating the quality of fields-of-view in bright-field microscope images of cervical cells. The approach consists of the construction of neighbourhood graphs using the nuclei as the set of vertices. Transformations are then applied to such graphs in order to highlight the main structures in the image. The methods result in the delineation of regions with varying cell density and the identification of cell clusters. Clustering methods are evaluated using a dataset of manually delineated clusters and compared to a related work.

Paper Nr: 84
Title:

Prediction of Organ Geometry from Demographic and Anthropometric Data based on Supervised Learning Approach using Statistical Shape Atlas

Authors:

Yoshito Otake, Carneal Catherine, Blake Lucas, Gaurav Thawait, John Carrino, Brian Corner, Marina Carboni, Barry DeCristofano, Michale Maffeo, Andrew Merkle and Mehran Armand

Abstract: We propose a method relating internal human organ geometries and non-invasively acquired information such as demographic and anthropometric data. We first apply a dimensionality reduction technique to a training dataset to represent the organ geometry with low dimensional feature coordinates. Regression analysis is then used to determine a regression function between feature coordinates and the external measurements of the subjects. Feature coordinates for the organ of an unknown subject are then predicted from external measurements using the regression function, subsequently the organ geometry is estimated from the feature coordinates. As an example case, lung shapes represented as a point distribution model was analyzed based on demographic (age, gender, race), and several anthropometric measurements (height, weight, and chest dimensions). The training dataset consisted of 124 topologically consistent lung shapes created from thoracic CT scans. The prediction error of lung shape of an unknown subject based on 11 demographic and anthropometric information was 10.71 ± 5.48 mm. This proposed approach is applicable to scenarios where the prediction of internal geometries from external parameters is of interest. Examples include the use of external measurements as a prior information for image quality improvement in low dose CT, and optimization of CT scanning protocol.

Paper Nr: 85
Title:

Tracking Subpixel Targets with Critically Sampled Optics

Authors:

James Lotspeich and Mathias Kolsch

Abstract: In many remote sensing applications, the area of a scene sensed by a single pixel can often be measured in squared meters. This means that many objects of interest in a scene are smaller than a single pixel in the resulting image. Current tracking methods rely on robust object detection using multi-pixel features. A subpixel object does not provide enough information for these methods to work. This paper presents a method for tracking subpixel objects in image sequences captured from a stationary sensor that is critically sampled. Using template matching, we make a Maximum a Posteriori estimate of the target state over a sequence of images. A distance transform is used to calculate the motion prior in linear time, dramatically decreasing computation requirements. We compare the results of this method to a track-before-detect particle filter designed for tracking small, low contrast objects using both synthetic and real-world imagery. Results show our method produces more accurate state estimates and higher detection rates than the current state of the art methods at signal-to-noise ratios as low as 3dB.

Paper Nr: 108
Title:

Explaining Unintelligible Words by Means of their Context

Authors:

Balázs Pintér, Gyula Vörös, Zoltán Szabó and András Lőrincz

Abstract: Explaining unintelligible words is a practical problem for text obtained by optical character recognition, from the Web (e.g., because of misspellings), etc. Approaches to wikification, to enriching text by linking words to Wikipedia articles, could help solve this problem. However, existing methods for wikification assume that the text is correct, so they are not capable of wikifying erroneous text. Because of errors, the problem of disambiguation (identifying the appropriate article to link to) becomes large-scale: as the word to be disambiguated is unknown, the article to link to has to be selected from among hundreds, maybe thousands of candidate articles. Existing approaches for the case where the word is known build upon the distributional hypothesis: words that occur in the same contexts tend to have similar meanings. The increased number of candidate articles makes the difficulty of spuriously similar contexts (when two contexts are similar but belong to different articles) more severe. We propose a method to overcome this difficulty by combining the distributional hypothesis with structured sparsity, a rapidly expanding area of research. Empirically, our approach based on structured sparsity compares favorably to various traditional classification methods.

Paper Nr: 113
Title:

Measuring Linearity of Curves

Authors:

Joviša Žunić, Jovanka Pantović and Paul L. Rosin

Abstract: In this paper we define a new linearity measure which can be applied to open curve segments. The new measure ranges over the interval (0;1]; and produces the value 1 if and only if the measured line is a perfect straight line segment. Also, the new linearity measure is invariant with respect to translations, rotations and scaling transformations.

Paper Nr: 127
Title:

Video Segmentation based on Multi-kernel Learning and Feature Relevance Analysis for Object Classification

Authors:

S. Molina-Giraldo, J. Carvajal-González, A. M. Álvarez-Meza and G. Castellanos-Domínguez

Abstract: A methodology to automatically detect moving objects in a scene using static cameras is proposed. Using Multiple Kernel Representations, we aim to incorporate multiple information sources in the process, and employing a relevance analysis, each source is automatically weighted. A tuned Kmeans technique is employed to group pixels as static or moving objects. Moreover, the proposed methodology is tested for the classification of abbandoned objects. Attained results over real-world datasets, show how our approach is stable using the same parameters for all experiments.

Short Papers
Paper Nr: 8
Title:

Automatic Update and Completion of Occluded Regions for Accurate 3D Urban Cartography by combining Multiple Views and Multiple Passages

Authors:

Ahmad Kamal Aijazi, Paul Checchin and Laurent Trassoudaine

Abstract: Handling occlusions is one of the more difficult challenges faced today in urban landscape analysis and cartography. In this paper, we successfully address this problem by using a new method in which multiple views and multiple sessions or passages are used to complete occluded regions in a 3D cartographic map. Two 3D point clouds, from different viewing angles, obtained in each passage are first classified into two main object classes: Permanent and Temporary (which contains both Temporarily static and Mobile objects) using inference based on basic reasoning. All these Temporary objects, considered as occluding objects, are removed from the scene leaving behind two perforated 3D point clouds of the cartography. These two perforated point clouds from the same passage are then combined together to fill in some of the holes and form a unified perforated 3D point cloud of the cartography. This unified perforated 3D point cloud is then updated by similar subsequent perforated point clouds, obtained on different days and hours of the day, filling in the remaining holes and completing the missing features/regions of the urban cartography. This automatic method ensures that the resulting 3D point cloud of the cartography is most accurate containing only the exact and actual permanent features/regions. Special update and reset functions are added to increase the robustness of the method. The method is evaluated on a standard data set to demonstrate its efficacy and prowess.

Paper Nr: 21
Title:

A Search Engine for Retrieval and Inspection of Events with 48 Human Actions in Realistic Videos

Authors:

G. J. Burghouts, L. de Penning, M. Kruithof, P. Hanckmann, J-M Ten Hove, S. Landsmeer, S. P. van den Broek, R. den Hollander, C. van Leeuwen, S. Korzec, H. Bouma and K. Schutte

Abstract: The contribution of this paper is a search engine that recognizes and describes 48 human actions in realistic videos. The core algorithms have been published recently, from the early visual processing (Bouma, 2012), discriminative recognition (Burghouts, 2012) and textual description (Hanckmann, 2012) of 48 human actions. We summarize the key algorithms and specify their performance. The novelty of this paper is that we integrate these algorithms into a search engine. In this paper, we add an algorithm that finds the relevant spatio-temporal regions in the video, which is the input for the early visual processing. As a result, meta-data is produced by the recognition and description algorithms. The meta-data is filtered by a novel algorithm that selects only the most informative parts of the video. We demonstrate the power of our search engine by retrieving relevant parts of the video based on three different queries. The search results indicate where specific events occurred, and which actors and objects were involved. We show that events can be successfully retrieved and inspected by usage of the proposed search engine.

Paper Nr: 26
Title:

Investigating Feature Extraction for Domain Adaptation in Remote Sensing Image Classification

Authors:

Giona Matasci, Lorenzo Bruzzone, Michele Volpi, Devis Tuia and Mikhail Kanevski

Abstract: In this contribution, we explore the feature extraction framework to ease the knowledge transfer in the thematic classification of multiple remotely sensed images. By projecting the images in a common feature space, the purpose is to statistically align a given target image to another source image of the same type for which we dispose of already collected ground truth. Therefore, a classifier trained on the source image can directly be applied on the target image. We analyze and compare the performance of classic feature extraction techniques and that of a dedicated method issued from the field of domain adaptation. We also test the influence of different setups of the problem, namely the application of histogram matching and the origin of the samples used to compute the projections. Experiments on multi- and hyper-spectral images reveal the benefits of the feature extraction step and highlight insightful properties of the different adopted strategies.

Paper Nr: 33
Title:

Minimal Structure and Motion Problems for TOA and TDOA Measurements with Collinearity Constraints

Authors:

Erik Ask, Simon Burgess and Karl Åström

Abstract: Structure from sound can be phrased as the problem of determining the position of a number of microphones and a number of sound sources given only the recorded sounds. In this paper we study minimal structure from sound problems in both TOA (time of arrival) and TDOA (time difference of arrival) settings with collinear constraints on e.g. the microphone positions. Three such minimal cases are analyzed and solved with efficient and numerically stable techniques. An experimental validation of the solvers are performed on both simulated and real data. In the paper we also show how such solvers can be utilized in a RANSAC framework to perform robust matching of sound features and then used as initial estimates in a robust non-linear leastsquares optimization.

Paper Nr: 47
Title:

Baseline Estimation in Arabic Handwritten Text-Line - Evaluation on AHTID/MW Database

Authors:

Anis Mezghani, Slim Kanoun, Souhir Bouaziz, Maher Khemakhem and Haikal El Abed

Abstract: Baseline extraction is one of the most important phases for handwriting recognition. Due to the complexity of the Arabic scripts, baseline detection of Arabic handwritten text-lines is a difficult task compared to other languages. In this work, a method which combines some baseline extraction techniques used in literature was presented to provide a fine estimation of baseline in Arabic handwritten text-lines. For evaluation purpose, the AHTID/MW database was extended by a baseline ground truth annotation. The database is freely available for researchers worldwide which enable other researchers to test their baseline detection systems.

Paper Nr: 78
Title:

Arabic Corpus Enhancement using a New Lexicon/Stemming Algorithm

Authors:

Ashraf AbdelRaouf, Colin A. Higgins, Tony Pridmore and Mahmoud I. Khalil

Abstract: Optical Character Recognition (OCR) is an important technology and has many advantages in storing information for both old and new documents. The Arabic language lacks both the variety of OCR systems and the depth of research relative to Roman scripts. An authoritative corpus is beneficial in the design and construction of any OCR system. Lexicon and stemming tools are essential in enhancing corpus retrieval and performance in an OCR context. A new lexicon/stemming algorithm is presented based on the Viterbi path method which uses a light stemmer approach. Lexicon and stemming lookup is combined to obtain a list of alternatives for uncertain words. This list removes affixes (prefixes or suffices) if there are any; otherwise affixes are added. Finally, every word in the list of alternatives is verified by searching the original corpus. The lexicon/stemming algorithm also assures the continuous updating of the contents of the corpus presented by (AbdelRaouf et al., 2010), which copes with the innovative needs of Arabic OCR research.

Paper Nr: 87
Title:

Experimental Evaluation of Probabilistic Similarity for Spoken Term Detection

Authors:

Shi-wook Lee, Hiroaki Kojima, Kazuyo Tanaka and Yoshiaki Itoh

Abstract: In this paper, the use of probabilistic similarity and the likelihood ratio for spoken term detection is investigated. The object of spoken term detection is to rank retrieved spoken terms according to their distance from a query. First, we evaluate several probabilistic similarity functions for use as a sophisticated distance. In particular, we investigate probabilistic similarity for Gaussian mixture models using the closed-form solutions and pseudo-sampling approximation of Kullback–Leibler divergence. And then we propose additive scoring factors based on the likelihood ratio of each individual subword. An experimental evaluation demonstrates that we can achieve an improved detection performance by using probabilistic similarity functions and applying the likelihood ratio.

Paper Nr: 88
Title:

Handling Unbalanced Data in Nocturnal Epileptic Seizure Detection using Accelerometers

Authors:

Kris Cuppens, Peter Karsmakers, Anouk Van de Vel, Bert Bonroy, Milica Milosevic, Lieven Lagae, Berten Ceulemans, Sabine Van Huffel and Bart Vanrumste

Abstract: Data of nocturnal movements in epileptic patients is marked by an imbalance due to the relative small number of seizures compared to normal nocturnal movements. This makes developing a robust classifier more difficult, especially with respect to reducing the number of false positives while keeping a high sensitivity. In this paper we evaluated different ways to overcome this problem in our application, by using a different weighting of classes and by resampling the minority class. Furthermore, as we only have a limited number of training samples available per patient, additionally it was investigated in which manner the training set size affects the results. We observed that oversampling gives a higher performance than only adjusting the weights of both classes. Compared to its alternatives oversampling based on the probability density function gives the best results. On 2 of 3 patients, this technique gives a sensitivity of 95% or more and a PPV more than 70%. Furthermore, an increased imbalance in the dataset leads to lower performance, whereas the size of the dataset has little influence.

Paper Nr: 89
Title:

Reduced Search Space for Rapid Bicycle Detection

Authors:

M. Nilsson, H. Ardö, A. Laureshyn and A. Persson

Abstract: This paper describes a solution to the application of rapid detection of bicycles in low resolution video. In particular, the application addressed is from video recorded in a live environment. The future aim from the results in this paper is to investigate a full year of video data. Hence, processing speed is of great concern. The proposed solution involves the use of an object detector and a search space reduction method based on prior knowledge regarding the application at hand. The method using prior knowledge utilizes random sample consensus, and additional statistical analysis on detection outputs, in order to define a reduced search space. It is experimentally shown that, in the application addressed, it is possible to reduce the full search space by 62% with the proposed methodology. This approach, which employs a full detector in combination with the design of a simple and fast model that can capture prior knowledge for a specific application, leads to a reduced search space and thereby a significantly improved processing speed.

Paper Nr: 92
Title:

Droplet Tracking from Unsynchronized Cameras

Authors:

L. A. Zarrabeitia, F. Z. Qureshi and D. A. Aruliah

Abstract: We develop a method for reconstructing the three-dimensional (3D) trajectories of droplets in flight captured by multiple unsynchronized cameras. Triangulation techniques that underpin most stereo reconstruction methods do not apply in this context as the image streams recorded by different cameras are not synchronized. We assume instead a priori knowledge about the motion of the droplets to reconstruct their 3D trajectories given unlabelled two-dimensional tracks in videos recorded by different cameras. Our approach also avoids the challenge of accurately matching droplet tracks across multiple video frames. We evaluate the proposed method using both synthetic and real data.

Paper Nr: 93
Title:

4D Polygonal Approximation of the Skeleton for 3D Object Decomposition

Authors:

Luca Serino, Carlo Arcelli and Gabriella Sanniti di Baja

Abstract: We improve a method to decompose a 3D object into parts (called kernels, simple-regions and bumps) starting from the partition of the distance labeled skeleton into components (called complex-sets, simple-curves and single-points). In particular, each simple-curve of the partition is here interpreted as a curve in a 4D space, where the coordinates of each point are related to the three spatial coordinates of the corresponding voxel of the 3D simple-curve and to its associated distance label. Then, a split type polygonal approximation method is employed to subdivide, in the limits of the adopted tolerance, each curve in the 4D space into straight-line segments. Vertices found in the 4D curve are used to identify corresponding vertices in the 3D simple-curve. The skeleton partition is then used to recover the parts into which the object is decomposed. Finally, region merging is taken into account to obtain a decomposition of the object more in accordance with human intuition.

Paper Nr: 103
Title:

Shape Representation via the Generalized Geodesic Median Point

Authors:

Foteini Fotopoulou, Dimitrios Kastaniotis, Ilias Theodorakopoulos and George Economou

Abstract: In this paper the concept of shape boundary description based on a stable point is revisited. In an effort to enhance the description based on approaches that observe the shape’s boundary from a stable point positioned in the inside shape area, we propose methods to define new such center points. Looking to the shape description from this point of view is not just a new shape signature but new possibilities are provided due to the combination of the different descriptions. The produced 1D sequences could be processed along their time evolution in scalar or vector (multichannel) fashion and enhance the discriminative capability of the combined features a great deal. The new center point proposed here is the generalized Geodesic Median center-point and it is produced as the minimum of a distance function, using Geodesic distances to achieve bending invariance - a property which is not possessed by Euclidean distances. Experiments on the MPEG7 and the Kimia99 shape databases reveal the effectiveness of the introduced method against other, of higher computational complexity, state-of-art methods that are also based on geodesic distances.

Paper Nr: 115
Title:

Evidence Accumulation Approach applied to EEG Analysis

Authors:

Helena Aidos, Carlos Carreiras, Hugo Silva and Ana Fred

Abstract: Human-machine interaction is a rapidly expanding field which benefits from automatic emotion recognition. Therefore, methods that can automatically detect the emotional state of a person are important for this field, as well as for fields such as psychology and psychiatry. This paper proposes the use of clustering ensembles (CEs) to achieve such detection. We use CEs on a dataset containing EEG signals from subjects who performed a stress-inducing task. From the raw EEG data we apply filtering and processing techniques leading to three dataset types: simple EEG, EEG with eye-movement artifacts removed through Independent Component Analysis, and data-driven modes extracted using Empirical Mode Decomposition. Then, for each of these three data types, we compute band power features and phase-locking factors, yielding a total of six different feature spaces. These spaces are then analyzed using the CE framework which combines results of multiple clustering algorithms in a voting scheme. This procedure yields interesting clusters, in particular a natural tendency for finding low numbers of clusters per subject and finding clusters which are composed of consecutive test lines. These two facts combined may indicate that a change in the emotional state of the subject was detected by the proposed framework.

Paper Nr: 155
Title:

Performance Evaluation of Image Filtering for Classification and Retrieval

Authors:

Falk Schubert and Krystian Mikolajczyk

Abstract: Much research effort in the literature is focused on improving feature extraction methods to boost the performance in various computer vision applications. This is mostly achieved by tailoring feature extraction methods to specific tasks. For instance, for the task of object detection often new features are designed that are even more robust to natural variations of a certain object class and yet discriminative enough to achieve high precision. This focus led to a vast amount of different feature extraction methods with more or less consistent performance across different applications. Instead of fine-tuning or re-designing new features to further increase performance we want to motivate the use of image filters for pre-processing. We therefore present a performance evaluation of numerous existing image enhancement techniques which help to increase performance of already well-known feature extraction methods. We investigate the impact of such image enhancement or filtering techniques on two state-of-the-art image classification and retrieval approaches. For classification we evaluate using a standard Pascal VOC dataset. For retrieval we provide a new challenging dataset. We find that gradient-based interest-point detectors and descriptors such as SIFT or HOG can benefit from enhancement methods and lead to improved performance.

Paper Nr: 158
Title:

Stereo-based Spatial and Temporal Feature Matching Method for Object Tracking and Distance Estimation

Authors:

Young-Chul Lim, Chung-Hee Lee and Jonghwan Kim

Abstract: In this paper, we propose a stereo-based object tracking and distance estimation method using spatial and temporal feature matching scheme. Our work aims to track an object robustly and to estimate its distance accurately without stereo matching processing, which requires a considerable amount of processing time and numerous processing resources. Our method combines temporal feature matching and spatial feature matching schemes. Our experimental results demonstrate that the proposed method can provide good object tracking and distance estimation performance in real-world environments.

Paper Nr: 160
Title:

Rotated Local Binary Pattern (RLBP) - Rotation Invariant Texture Descriptor

Authors:

Rakesh Mehta and Karen Egiazarian

Abstract: In this paper we propose two novel rotation invariant local texture descriptors. They are based on Local Binary Pattern (LBP), which is one of the most effective and frequently used texture descriptor. Although LBP efficiently captures the local structure, it is not rotation invariant. In the proposed methods, a dominant direction is evaluated in a circular neighbourhood and the descriptor is computed with respect to it. The weights associated with the neighbouring pixels are circularly shifted with respect to this dominant direction. Further, in the second descriptor, the uniformity of the patterns is utilized to extract more discriminative information. The proposed methods are tested for the task of texture classification and the performance is compared with original LBP and its existed extensions.

Paper Nr: 161
Title:

Integrating MicroRNA and mRNA Expression Data for Cancer Classification

Authors:

Hasan Oğul and Onur Altındağ

Abstract: Classifying cancer samples from gene expression data is one of the central problems in current systems biomedicine. The problem is challenging due to the small number of samples in comparison to the number of genes (mRNAs) in a typical microarray experiment. Recent reports suggest that feature selection may help to manage the problem. Furthermore, microRNA expression profiles have shown to provide valuable knowledge in detecting cancer signatures. In this study, we present the results of a comprehensive study to assess the effect of feature selection and microRNA-mRNA data integration in cancer type prediction from microarray expression data. We prove that this integration can significantly improve prediction accuracy with a proper feature selection strategy.

Posters
Paper Nr: 20
Title:

Recognition of Untrustworthy Face Images in ATM Sessions using a Bio-inspired Intelligent Network

Authors:

R. Škoviera, K. Valentín, S. Štolc and I. Bajla

Abstract: The aim of this paper is to report on a pilot application of a bio-inspired intelligent network model, called Hierarchical Temporal Memory (HTM), for recognition (detection) of untrustworthy manipulation with an Automatic Teller Machine (ATM). HTM was used as a crucial part of an anomaly detection system to recognize hard-to-identifiable faces, i.e., faces with a mask, covered with a scarf, or wearing sunglasses. Those types of face occlusion can be a good indicator of potentialy malicious intentions of an ATM user. In the presented system, the Kinect camera was used for acquisition of video image sequences. The Kinect’s depth output along with skeleton tracking was used as a basis of the color image segmentation. To test the proposed system, experiments have been carried out in which several participants performed normal and untrustworthy actions using an ATM simulator. The output of the face classification system can assist a security personnel in surveillance tasks.

Paper Nr: 27
Title:

Unsupervised Light Spot Detection using Background Subtraction

Authors:

Takaya Niwa and Kazuhiro Hotta

Abstract: Live cell imaging has been developing rapidly by the development of the microscope and fluorescence technique. Light spot detection in intracellular image is important for elucidation of form of morphology of animal. However, light spots are detected manually now, and human can not treat a large number of images. If automatic detection by computer is realized, we can obtain many objective data, and it will be useful for the biology development. In general, supervised learning is useful to develop a good detector. However, many particles are included in an intracellular image, and it is difficult to make a lot of supervised samples. Therefore, in this paper, we propose a light spot detection method based on unsupervised learning. Concretely, we use background subtraction and robust statistics to detect the light spots. In experiments using Wnt-3a images, the proposed method outperforms ImageJ which is usually used in cell biology.

Paper Nr: 38
Title:

Research of Classification Algorithms for Recognition of Digits in Mechanical Counters

Authors:

Dirk Benyoucef, Pirmin Held and Philipp Klein

Abstract: Mechanical counters are still very popular for their protection against manipulation and low costs. In the past automatic readout of mechanical counters required complex and expensive image processing methods. The system proposed in this paper is a cheaper alternative which does not require modifications to the mechanics of the counter. The proposed system makes use of different light reflectivity parameters of the numbers shown on the number wheels. In this paper the different approaches are shown and analyzed.

Paper Nr: 43
Title:

Symbolic Representation over Skeleton Endpoints for Classification of Medical X-ray Images

Authors:

Amir Rajaei, Elham Dallalzadeh and Lalitha Rangarajan

Abstract: In this paper, we propose a model for symbolic representation and classification of medical X-ray body organ images. Medical X-ray body organ images are segmented using graph cut segmentation method. Based on the boundary of a segmented body organ image, the skeleton endpoints are localized. A complete directed graph is then constructed over the skeleton endpoints. Subsequently, distance and orientation features are extracted from the constructed graph. Further, shape features based on skeleton endpoints are also extracted. The obtained features are used to form an interval valued feature vector. Finally, a symbolic classifier is explored to classify medical X-ray body organ images. Our proposed model is simple and efficient.

Paper Nr: 55
Title:

Enhancing Security Event Management Systems with Unsupervised Anomaly Detection

Authors:

Markus Goldstein, Stefan Asanger, Matthias Reif and Andrew Hutchison

Abstract: Security Information and Event Management (SIEM) systems are today a key component of complex enterprise networks. They usually aggregate and correlate events from different machines and perform a rule-based analysis to detect threats. In this paper we present an enhancement of such systems which makes use of unsupervised anomaly detection algorithms without the need for any prior training of the system. For data acquisition, events are exported from an existing SIEM appliance, parsed, unified and preprocessed to fit the requirements of unsupervised anomaly detection algorithms. Six different algorithms are evaluated qualitatively and finally a global k-NN approach was selected for a practical deployment. The new system was able to detect misconfigurations and gave the security operation center team more insight about processes in the network.

Paper Nr: 57
Title:

Rule-based Hand Posture Recognition using Qualitative Finger Configurations Acquired with the Kinect

Authors:

Lieven Billiet, Jose Oramas, McElory Hoffmann, Wannes Meert and Laura Antanas

Abstract: Gesture recognition systems exhibit failures when faced with large hand posture vocabularies or relatively new hand poses. One main reason is that 2D and 3D appearance-based approaches require significant amounts of training data. We address this problem by introducing a new 2D model-based approach to recognize hand postures. The model captures a high-level rule-based representation of the hand expressed in terms of finger poses and their qualitative configuration. The available 3D information is used to first segment the hand. We evaluate our approach on a Kinect dataset and report superior results while using less training data when comparing to state-of-the-art 3D SURF descriptor.

Paper Nr: 69
Title:

Deriving Basic Law of Human Mobility using Community - Contributed Multimedia Data

Authors:

Katarina Gavrić, Dubravko Ćulibrk and Vladimir Crnojević

Abstract: In recent years, geo-referenced community-contributed multimedia data that is available from services such as Flickr/YouTube, has been used to help understand patterns of human mobility, behavior and habits. While this data is freely available for much larger regions of the world, it is understood that the quality of such data is lower than that of data that can be obtained from mobile phone operators. This is probably the reason why public data has not been considered for studies attempting to identify basic laws that govern human mobility. In this study we explore the possibility of using Flickr data as an alternative to mobile-phone-generated data when it comes to analyzing human mobility. To do this, we apply a recently published approach to analysis of mobile phone data to the trajectories of 6404 Fickr users, derived from a dataset of 1 million images pertinent to the San Francisco/San Diego area. Our goal is to show that regularities that can be observed using mobile phone data are present in the Flickr data and that the publicly available data has the potential to enable researchers to conduct similar analysis at larger (continent/world wide) scales, with possible applications to urban planning, traffic forecasting and the spread of biological and mobile-phone viruses. The results presented show that Flickr data is suitable for such studies, and can be used as an alternative to proprietary mobile-phone-use related data.

Paper Nr: 81
Title:

Robust Object Segmentation using Active Contours and Shape Prior

Authors:

Mohamed Amine Mezghich, Malek Sellami, Slim M'hiri and Faouzi Ghorbel

Abstract: In this paper, we intend to present new method to incorporate geometric shape prior into region-based active contours in order to improve its robustness to noise and occlusions. The proposed shape prior is defined after the registration of binary images associated with level set functions of the active contour and a reference shape. The used registration method is based on phase correlation. This representation makes it possible to manage several objects simultaneously. Experimental results show the ability of the proposed geometric shape prior to constrain an evolving curve towards a target shape. We highlight, on synthetic and real images, the benefit of the method on segmentation results in presence of partial occlusions, low contrast and noise.

Paper Nr: 83
Title:

Non-local Huber Regularization for Image Denoising - A Hybrid Approach of Two Non-local Regularizations

Authors:

Suil Son, Deokyoung Kang and Suk I. Yoo

Abstract: Non-local Huber regularization is proposed for image denoising. This method improves the non-local total variation regularization and the non-local H1 regularization approaches. The non-local total variation regularization preserves edges better than the non-local H1 regularization; however, it leaves a little noise. In contrast, the non-local H1 regularization eliminates noise better than the non-local total variation regularization; however, it blurs edges. To take both advantages of the two methods, the proposed method applies the non-local total variation to large non-local intensity differences and applies the non-local H1 regularization to small non-local intensity differences. A boundary value to determine whether the intensity difference comes from edges or noise is also suggested. The experimental results of the proposed method is compared to the result from the non-local total variation regularization and to the result from the non-local H1 regularization; The effect of the boundary value is illustrated as PSNR changes with respect to the various values of the boundary values.

Paper Nr: 102
Title:

Entropy based Biometric Template Clustering

Authors:

Michele Nappi, Daniel Riccio and Maria De Marsico

Abstract: Though speed and accuracy are two competing requirements for large scale biometric recognition, they both suffer from large database size. Clustering seems promising to reduce the search space. This can improve accuracy, but may even contrarily affect it by a poor selection of the candidate cluster for the search. We present a novel technique that exploits gallery entropy for clustering. The comparison with K-Means demonstrates that we achieve a better clustering result, yet without fixing the number of clusters a-priori.

Paper Nr: 104
Title:

Page Analysis by 2D Conditional Random Fields

Authors:

Atsuhiro Takasu

Abstract: This paper applies two-dimensional conditional random fields (2D CRF) to page analysis and information extraction. In this paper we discuss features and labels for information extraction by 2D CRF. We evaluated the method by applying it to the problem of extracting bibliographic components from scanned title pages of academic papers. The experimental results show that 2D CRF improves the performance of information extraction compared to chain-model CRF.

Paper Nr: 111
Title:

A Combined SVM/HCRF Model for Activity Recognition based on STIPs Trajectories

Authors:

Mouna Selmi, Mounim A. El-Yacoubi and Bernadette Dorizzi

Abstract: In this paper, we propose a novel human activity recognition approach based on STIPs’ trajectories as local descriptors of video sequences. This representation compares favorably with state of art feature extraction methods. In addition, we investigate the use of SVM/HCRF combination for temporal sequence modeling, where SVM is applied locally on short video segments to produce probability scores, the latter being considered as the input vectors to HCRF. This method constitutes a new contribution to the state of the art on activity recognition task. The obtained results demonstrate that our method is efficient and compares favorably with state of the art methods on human activity recognition.

Paper Nr: 123
Title:

Semi-supervised Discovery of Time-series Templates for Gesture Spotting in Activity Recognition

Authors:

Héctor F. Satizábal, Julien Rebetez and Andres Perez-Uribe

Abstract: In human activity recognition, gesture spotting can be achieved by comparing the data from on-body sensors with a set of known gesture templates. This work presents a semi-supervised approach to template discovery in which the Dynamic Time Warping distance measure has been embedded in a classic clustering technique. Clustering is used to find a set of template candidates in an unsupervised manner, which are then evaluated by means of a supervised assessment of their classification performance. A cross-validation test over a benchmark dataset showed that our approach yields good results with the advantage of using a single sensor.

Paper Nr: 124
Title:

An Information Theoretic Approach to Text Sentiment Analysis

Authors:

David Pereira Coutinho and Mário A. T. Figueiredo

Abstract: Most approaches to text sentiment analysis rely on human generated lexicon-based feature selection methods, supervised vector-based learning methods, and other solutions that seek to capture sentiment information. Most of these methods, in order to yield acceptable accuracy, require a complex preprocessing stage and careful feature engineering. This paper introduces a coding-theoretic-based sentiment analysis method that dispenses with any text preprocessing or explicit feature engineering, but still achieves state-of-the-art accuracy. By applying the Ziv-Merhav method to estimate the relative entropy (Kullback-Leibler divergence) and the cross parsing length from pairs of sequences of text symbols, we get information theoretic measures that make very few assumptions about the models which are assumed to have generated the sequences. Using these measures, we follow a dissimilarity space approach, on which we apply a standard support vector machine classifier. Experimental evaluation of the proposed approach on a text sentiment analysis problem (more specifically, movie reviews sentiment polarity classification) reveals that it outperforms the previous state-of-the-art, despite being much simpler than the competing methods.

Paper Nr: 131
Title:

Smart Classifier Selection for Activity Recognition on Wearable Devices

Authors:

Negar Ghourchian and Doina Precup

Abstract: Activity recognition is a key component of human-machine interaction applications. Information obtained from sensors in smart wearable devices is especially valuable, because these devices have become ubiquitous, and they record large amounts of data. Machine learning algorithms can then be used to process this data. However, wearable devices impose restrictions in terms of computation and energy resources, which need to be taken into account by a learning algorithm. We propose to use a real-time learning approach, which interactively determines the most effective set of modalities (or features) for classification, given the task at hand. Our algorithm optimizes sensor selection, in order to consume less power, while still maintaining good accuracy in classifying sequences of activities. Performance on a large, noisy dataset including four different sensing modalities shows that this is a promising approach.

Paper Nr: 132
Title:

Grid-based Spatial Keypoint Selection for Real Time Visual Odometry

Authors:

Volker Nannen and Gabriel Oliver

Abstract: Robotic systems can achieve real-time visual odometry by extracting a fixed number of invariant keypoints from the current camera frame, matching them against keypoints from a previous frame, and calculating camera motion from matching pairs. If keypoints are selected by response only they can become concentrated in a small image region. This decreases the chance for keypoints to match between images and increases the chance for a degenerate set of matching keypoints. Here we present and evaluate a simple grid-based method that forces extracted keypoints to follow an even spatial distribution. The benefits of this approach depend on image quality. Real world trials with low quality images show that the method can extend the length of a correctly estimated path by an order of magnitude. In laboratory trials with images of higher quality we observe that the quality of motion estimates can degrade significantly, in particular if the number of extracted keypoints is low. This negative effect can be minimized by using a large number of grid cells.

Paper Nr: 135
Title:

Handwritten Basic Word Recognition System based on Statistical-structural Features and HMM-treatment

Authors:

G. Facchini

Abstract: This paper presents a new approach for handwritten word recognition, based on fusion of statistics and structural features by sliding windows in a HMM (Hidden Markov Models) based system. In this new approach, structural features are encoded in a statistical representation unique and appended to the other statistical features to calculate the new type of feature. Experimental results obtained for basic word recognition in legal amount treatment are showing that the approach allows a increasing of the recognition rate. The experimental results are reported in the paper.

Paper Nr: 150
Title:

Measuring Musical Rhythm Similarity - Statistical Features versus Transformation Methods

Authors:

Juan Felipe Beltran, Xiaohua Liu, Nishant Mohanchandra and Godfried Toussaint

Abstract: Two approaches to measuring the similarity between symbolically notated musical rhythms are compared with human judgments of perceived similarity. The first is the edit-distance, a popular transformation method, applied to the rhythm sequences. The second works on the histograms of the inter-onset-intervals (IOIs) of these rhythm sequences. Furthermore, two methods of dealing with the histograms are also compared: the Mallows distance, and the employment of a group of standard statistical features. The results provide further evidence from the aural domain, that transformation methods are superior to feature-based methods for predicting human judgments of similarity. Furthermore, the results also support the hypothesis that statistical features applied to the histograms of the rhythms are better than music-theoretical structural features applied to the rhythms themselves.

Paper Nr: 152
Title:

The Influence of the Sampling Frequency of a Video Recording on a Heart-rate Detection Algorithm

Authors:

Marina C. Cidota, Dragos Datcu and Leon J. M. Rothkrantz

Abstract: Contact-free technology has a great potential in different medical areas such as personal health monitoring and telemedicine. One of the physiological parameters that can be measured with this type of technology is the heart rate. The pulse is proportional with the physical effort or mental stress, its measurement being an important issue in sport, medicine and psychology. In this paper we present an analysis of the accuracy of the heart rate detection using a high speed camera for recording a color video with the face of a person. The recordings were done both from frontal view and from profile and they were resampled to different frequencies between 10 and 240 frames per second. From our tests we may conclude that it was not the high frequency but the quality of the videos recorded with a high speed camera that allowed us to reduce the time needed to obtain the heart-rate up to 5 seconds. We have also noticed that the results are influenced by the errors generated in the resampling process of the video signal.

Paper Nr: 157
Title:

Non-uniform Quantization of Detail Components in Wavelet Transformed Image for Lossy JPEG2000 Compression

Authors:

Madhur Srivastava and Prasanta K. Panigrahi

Abstract: The paper introduces the idea of non-uniform quantization in the detail components of wavelet transformed image. It argues that most of the coefficients of horizontal, vertical and diagonal components lie near to zeros and the coefficients representing large differences are few at the extreme ends of histogram. Therefore, this paper advocates need for variable step size quantization scheme which preserves the edge information at the edge of histogram and removes redundancy with the minimal number of quantized values. To support the idea, preliminary results are provided using a non-uniform quantization algorithm. We believe that successful implementation of non-uniform quantization in detail components in JPEG-2000 still image standard will improve image quality and compression efficiency with lesser number of quantized values.

Paper Nr: 159
Title:

Useful Pattern Mining on Time Series - Applications in the Stock Market

Authors:

Nikitas Goumatianos, Ioannis T. Christou and Peter Lindgren

Abstract: We present the architecture of a “useful pattern” mining system that is capable of detecting thousands of different candlestick sequence patterns at the tick or any higher granularity levels. The system architecture is highly distributed and performs most of its highly compute-intensive aggregation calculations as complex but efficient distributed SQL queries on the relational databases that store the time-series. We present initial results from mining all frequent candlestick sequences with the characteristic property that when they occur then, with an average at least 60% probability, they signal a 2% or higher increase (or, alternatively, decrease) in a chosen property of the stock (e.g. close-value) within a given time-window (e.g. 5 days). Initial results from a first prototype implementation of the architecture show that after training on a large set of stocks, the system is capable of finding a significant number of candlestick sequences whose output signals (measured against an unseen set of stocks) have predictive accuracy which varies between 60% and 95% depended on the type of pattern.

Paper Nr: 162
Title:

Image Labeling using Integration of Local and Global Features

Authors:

Takuto Omiya and Kazuhiro Hotta

Abstract: In this paper, we carry out image labeling based on probabilistic integration of local and global features. Many conventional methods put label to each pixel or region using the features extracted from local regions and local contextual relationships between neighboring regions. However, labeling results tend to depend on a local viewpoint. To overcome this problem, we propose the image labeling method using not only local features but also global features. We compute posterior probability of local and global features independently, and they are integrated by the product. To compute probability of global region (entire image), Bag-of-Words is used. On the other hand, local co-occurrence between color and texture features is used to compute local probability. In the experiments using MSRC21 dataset, labeling accuracy is much improved by using global viewpoint.