Adaptive Unithreshold Two-Bit Block Motion Estimation

Abstract

Image compression is essential technique to realize real time communication and motion estimation is the slowest part in the video processing. To reduce the complexity, one bit and two bit transform algorithms are presented past, but this algorithms are some problems, particularly when each pixel transform into binary, it increase encoding time. In this work, we improve the 2 bit transform algorithm and reduce the complexity while keeping subjective quality. When we do bit transforming, we compare the current block and previous frame’s collocated block. If the block is almost same, we just calculate current block's average and variance. If it is not, we find the mean and variance separately. By doing this method, we can approximately gain 10 % complexity decrease.