Class-Based Search Algorithm for Inter Mode Prediction of H.264/AVC

The H.264 provides various and useful features for improved coding efficiency and error robustness, but its encoder complexity is extremely high due to the computational time of ME (motion estimation) and mode decision of inter coding. In this paper, we propose a fast and simple class-based scheme to reduce the computational complexity of ME. The image in block base is classified to k classifications with SQCM (spatial-domain quickly classified method). Referring to the classified indices, we discriminate two methods of ME: one is pattern search for stationary macroblock, synchronous motion and unchanged object; other is fast inter prediction for instable or variant content of object. We can save the dispensable motion detection for different contents of video, named CQME (class-based quickly ME). Experimental results shows that the proposed method can achieve a reduction of 67% computational time on average, with an average PSNR loss of only 0.30 dB compared with the original H.264 reference software jm12.1.