Motion information based adaptive block classification for fast motion estimation

An object based fast search motion estimation algorithm is considered to be efficient since the computation of motion search can concentrate on blocks with moving object. However, the computational complexity of object segmentation is still too high to be applied on fast search motion estimation. In this paper, we propose an adaptive block classification algorithm. With this method, the fast search algorithm for motion estimation can make use of characteristics of motion information of the sequence being coded efficiently. In our approach, the statistical information which features the motion activities of the blocks in the previous frame is used to predict the characteristics of the motion activities of the blocks in the current frame. This is to re-assign the computation of motion search to locations that deserve to be searched more than others. Extensive experimental work has been done, results of which show that the adaptive block classification approach can accelerate the current fast search motion estimation algorithm with little decrease (0.01dB to 0.09dB for 8 standard test sequences) or sometimes, even increase (0.01dB to 0.17dB for 10 other standard test sequences) on resultant video quality, the peak signal-to-noise ratio (PSNR), while our fast algorithm is 150 times over the exhaustive full search algorithm on average.