CAS: Context Adaptive Search for motion estimation

Motion estimation is the key part of a video encoder. It helps to remove temporal redundancies in image sequences, but most motion estimation algorithms neglect to take the advantage of the string correlations within the motion fields. The search windows stay the same through the image sequences and the estimation needs heavy computation. To reduce this computational complexity, several algorithms have been proposed. One of the most popular methods is the logarithmic search. This method reduces the complexity but it neglects the high correlation of the motion field. Also, the number of search points increases as the search range increases. In this paper, we propose a fast motion estimation method that uses “context-adaptive search” (CAS) windows. It uses strong spatial correlation of the motion field and median prediction for motion-vector coding. The variable-size search window centred on the median values reduces the complexity of the motion vectors for motion-vector encoding. Simulation results show that the proposed method outperforms the logarithmic search in terms of compression efficiency by up to 8% with only one third of the search points. The overall peak SNR value is also slightly better