Distortion-Quantization Model And The Application In Mode Decision Algorithm For Video Coding

Abstract:Mode decision algorithm is a key technology in video encoder. Many video coding standards adopt mode decision based on rate distortion optimization to improve coding performance, such as H.264/AVC, AVS, and so on. However, this method needs to calculate R-D cost for all available modes to choose the best mode. Thus, the complexity increases quickly and it is hard for application. In this paper, we will focus on the issue above to improve rate-distortion optimized mode decision algorithm.According to the above ideas, the work in this paper is as follows:First, we introduce the development of video coding standard and the key technologies of video coding. The theory of rate distortion optimization and its application in video coding are also displayed in this paper. Especially, we focus on mode decision and show the weakness in the same time. In this paper, we will improve the mode decision by approximating R-D cost function and SKIP mode early decision.Next, we find that the model of DCT coefficients distribution have a direct impact on rate-distortion (D-Q) model. Thus, we have a simulation for several main DCT coefficients distribution models in this paper. Then, we have a fair comparison for mainstream D-Q models from accuracy, complexity and application. Based on above, we have an improvement for D-Q model to make it be used in rate distortion optimized mode decision.Finally, we find that SKIP mode accounts for a large proportion by testing a large num of video sequences. Currently, there have been some algorithms for SKIP mode early decision. Always, These algorithms are based on threshold judge. However, the algorithms can not be adaptive with video context very well. In this paper, we propose the context-based adaptive SKIP mode early decision (CASMED) algorithm to solve the issue above, we adjust the threshold by a parameter which can reflect the characteristic of current coding macroblock. In this paper, we adopt SATD(sum of absolute transformed difference) to adjust SAD(sum of absolute difference) threshold. Regression analysis and modeling offline are employed in our proposed algorithm. We have an analysis for large num of samples offline, then construct an accurate model for SAD and SATD. We can get the SAD threshold by putting the SATD of coding macroblock into the model.With the above methods using into AVS encoder, experimental results show that the proposed algorithm considerably reduce the computation complexity with only negligible coding performance loss.