New Application of Graph Mining to Video Analysis

Given a graph, frequent graph mining extracts subgraphs appearing frequently as useful knowledge. This paper proposes to exploit graph mining that discovers knowledge without supervision to realize unsupervised image analysis. In particular, we present a background subtraction algorithm from videos in which the background model is acquired without supervision. The targets of our algorithm are videos in which a moving object passes in front of a surveillance camera. After transforming each video frame into a region adjacency graph, our method discovers the subgraph representing the background, exploiting the fact that the background appears in more frames than the moving object.

[1]  Ming Yang,et al.  From frequent itemsets to semantically meaningful visual patterns , 2007, KDD '07.

[2]  Lawrence B. Holder,et al.  Substructure Discovery Using Minimum Description Length and Background Knowledge , 1993, J. Artif. Intell. Res..

[3]  Jung-Hwan Oh,et al.  STRG-Index: spatio-temporal region graph indexing for large video databases , 2005, SIGMOD '05.

[4]  Luc Van Gool,et al.  Video mining with frequent itemset configurations , 2006 .

[5]  Sebastian Nowozin,et al.  Weighted Substructure Mining for Image Analysis , 2007, 2007 IEEE Conference on Computer Vision and Pattern Recognition.

[6]  Jiawei Han,et al.  gSpan: graph-based substructure pattern mining , 2002, 2002 IEEE International Conference on Data Mining, 2002. Proceedings..

[7]  Wei-Ying Ma,et al.  Image and Video Retrieval , 2003, Lecture Notes in Computer Science.