DCPRAM 2015 Abstracts


Full Papers
Paper Nr: 1
Title:

Distributed Graph Matching and Graph Indexing Approaches - Applications to Pattern Recognition

Authors:

Zeina Abu-Aisheh, Romain Raveaux and Jean-Yves Ramel

Abstract: Attributed graphs are powerful data structures for the representation of complex objects. In a graph-based representation, vertices and their attributes describe objects (or part of objects) while edges represent interrelationships between the objects. Due to the inherent genericity of graph-based representations, and thanks to the improvement of computer capacities, structural representations have become more and more popular in the field of Pattern Recognition (PR). In this thesis, we tackle two important graph-based problems for PR: Graph Matching and Graph Indexing. The comparison between two objects is a crucial operation in PR. Representing objects by graphs turns the problem of object comparison into graph matching where correspondences between nodes and edges of two graphs have to be found. Moreover, graph-based indices are important so that a graph query can be retrieved from a large database via such indices, such a problem is referred to as graph indexing. The complexity of both graph matching and graph indexing is generally stated to be NP-COMPLETE or NP-hard. Coming up with a graph matching algorithm that can scale up to match graphs involved in PR tasks is a great challenge. Among the graph matching methods dedicated to PR problems, the Graph Edit Distance (GED) is of great interest. Over the last decade, GED has been applied to a wide range of specific applications from molecule recognition to image classification. In this report, we present the first part of the thesis. We tackle GED, shed light on the importance of having exact solutions rather than approximate ones and come up with a distributed GED where the search tree is decomposed into smaller trees which are solved independently and in a complete distributed manner. In the second part of the thesis, we aim at proposing new distributed graph-indexing approaches that aim at retrieving a graph from a large graph-based index as fast as possible. Graph indexing will be reported as a perspective of this work.

Paper Nr: 4
Title:

Web Usage Mining - MapReduce-based Emerging Pattern Mining in Hypergraph Learning

Authors:

Xiuming Yu, Meijing Li and Keun Ho Ryu

Abstract: Web usage mining is a popular research area in data mining. As the rapid growth of internet, more and more log information is collected by the web servers around the world. It becomes difficult to extract useful information from these huge web log data. Classic techniques of web usage mining are performed with low efficency in large number of web log data, because a lot of system resource is needed to deal with large computation. Most techniques of web mining are performed based on assosiation rule mining or frequent pattern mining, which aim to find relationships among web pages or predicting the behavoir of web users. That’s difficult to find some favourite web pages in different web users. In this research, we propose an efficient approach to find some favourite web pages in different web users in large web log data based on hypergraph learning by considering the programming model of MapReduce and the techniques of emerging pattern mining.

Paper Nr: 5
Title:

Collaborative Activities Understanding from 3D Data

Authors:

Fabrizio Natola, Valsamis Ntouksos and Fiora Pirri

Abstract: Our work consists in finding a way to recognize activities performed by two people that collaborate in a working environment. Starting from results obtained in the past years by Gong, Medioni and other authors, we go a step forward, trying to construct a learning function that is able to generalize the model provided by the authors cited before. Moreover, we search for a space in which we can map the points corresponding to the poses, over time, of the skeletons of the two subjects, so that no information is lost.

Paper Nr: 6
Title:

Human Activity Recognition and Prediction

Authors:

David Jardim, Luis Nunes and Miguel Sales Dias

Abstract: Human activity recognition (HAR) has become one of the most active research topics in image processing and pattern recognition. Detecting specific activities in a live feed or searching in video archives still relies almost completely on human resources. Detecting multiple activities in real-time video feeds is currently performed by assigning multiple analysts to simultaneously watch the same video stream. Manual analysis of video is labour intensive, fatiguing, and error prone. Solving the problem of recognizing human activities from video can lead to improvements in several applications fields like in surveillance systems, human computer interfaces, sports video analysis, digital shopping assistants, video retrieval, gaming and health-care. This area has grown dramatically in the past 10 years, and throughout our research we identified a potentially underexplored sub-area: Action Prediction. What if we could infer the future actions of people from visual input? We propose to expand the current vision-based activity analysis to a level where it is possible to predict the future actions executed by a subject. We are interested in interactions which can involve a single actor, two humans and/or simple objects. For example try to predict if “a person will cross the street” or “a person will try to steal a hand-bag from another” or were will a tenis-player target the next voley. Using a hierarchical approach we intend to represent high-level human activities that are composed of other simpler activities, which are usually called sub-events which may themselves be decomposable. We expect to develop a system capable of predicting the next action in a sequence initially using offline-learning and then with self-improvement/task specialization in mind, using online-learning.