Covariance analysis of non-translational motion-compensated frame differences

Abstract

Motion estimation is an important technique used in the compression coding of video data where significant temporal redundancy exists between neighboring video frames. Through motion estimation we can provide a motion-compensated prediction of the frame that is to be coded based on the previously coded frames. Coding the difference between the current frame and its motion-compensated prediction frame can be made more efficient if more accurate motion estimation can be achieved. Our previous work has shown that non-translation motion estimation based on Lie operators can increase the motion estimation accuracy by taking into account non-translation transformations of objects such as scaling and rotations across frames. As an extension of our previous work, we conduct covariance analysis on motion-compensated frame difference signals generated by (i) using conventional motion estimation methods that consider only translational motions, and (ii) applying additional non-translational motion estimation based on Lie operators. Experimental results on two standard test sequences show that our non-translational motion estimation method can further de-correlate the frame difference signals obtained by the conventional translational motion estimation methods.