Retrieval of objects in video by similarity based on graph matching

In this paper, we tackle the problem of matching of objects in video in the context of the rough indexing paradigm. The approach developed is based on matching of region adjacency graphs (RAG) of pre-segmented objects. In the context of the rough indexing paradigm, the video data are of very low resolution and segmentation is consequently inaccurate. Hence the RAGs vary with the time. The contribution of this paper is a graph matching method for such RAGs based on an improvement of relaxation labelling techniques. In this method, adjustments of similarity between regions according to neighborhood consistency compensate for the inaccuracy of segmentation. The approach demonstrates promising performance on real sequences when compared to another region-based technique.