Two Step Iterative Algorithm to Extract Generative Video Parameters from Video Sequences

GV (Generative Video) is a framework for the analysis and synthesis of video sequences. In GV, the operational units are not the actual frames in the original sequence; it has world images which have the non redundant information about the video sequences and the ancillary data. The world images and the ancillary data form the generative video representation, the information that is needed to regenerate the original video sequence. A two-step iterative algorithm is used here to obtain the generative video parameters. The first step estimates the background texture for a fixed template. The second step estimates the object template for a fixed background--the solution is given by a simple binary test evaluated at each pixel. The algorithm converges in a few iterations, typically three to five iterations. Keywords, Image/Video segmentation, Video compression, temporal integration.