Implications of a segmentation-based motion compensation scheme for the case of a block transform encoder

Abstract

Sipitca and Madisetti (see VCIP '96, Orlando, Florida, 1996) have proposed the use of analog position sensors for a more accurate motion compensation on videoconferencing sequences. We investigate the results of including such a motion compensation scheme in a transform based encoder. The more successful motion compensation makes the resulting displaced frame difference (DFD) less correlated. In fact, the autocorrelation function goes practically to zero for a lag larger than 4-5 pels. Therefore, the 4/spl times/4 DCT can be used instead of the 8/spl times/8 DCT. This helps in dealing better with the relatively small regions where the DFD takes on larger values-on the border regions and in the new information regions. Finally, a block swapping algorithm and an appropriate error measure are introduced.