Image Sequence Segmentation using a Single Evolutionary Graph Pyramid

In the presented method, an irregular pyramid is used to segment the successive frames of an image sequence: A pyramid is built with the first image of the sequence, and then it is updated from image to image, using a split-and-merge process that takes into account the changes occurred between two successive frames. Thus, the same pyramid structure is used along the sequence, speeding up the process. Homogeneity criteria allow to have the required compromise between speed and quality, i.e. to look for image evolution at a particular resolution. Thanks to the graph representation, objects obtained in a given image can be tracked in the rest of the sequence. 1 Introduction Second generation image coding techniques [3] allow to get rid of block effects, since they are based on objects instead of blocks. The irregular pyramid [5] [2] can be used to segment, or more generally to code an image, with the possibility of choosing a particular resolution at which the representation has to be do..