Mosaic generation in H.264 compressed domain

This paper presents an efficient approach to mosaic generation in H.264 compressed domain. The proposed mosaic generation approach fully exploits several new coding characteristics of the H.264 standard, i.e., variable block size and the quarter-pel accuracy of motion vector (MV). A uniformly sampled and normalized MVfield is first derived from the H.264 compressed video. Then an affine model is used to estimate the global motion from the normalized MV field, and the model parameters are iteratively refined based on outlier rejection scheme. Finally, the most recent and reliable background pixels from multiple frames are warped to the reference frame and then used to integrate the final mosaic. Experimental results demonstrate that the proposed approach is efficient for mosaic generation in H.264 compressed domain.