Complexity control for HEVC by visual saliency detection

The new quad-tree structure was adopt by High Efficiency Video Coding (HEVC) to partition the Coding Unit (CU). It improved compression efficiency while increased the computational complexity. This paper analysis the percentage of CU depth, the coding time of CU depth and the relationship between the CU depth and the visual saliency. A complexity control algorithm for HEVC with the visual saliency is proposed. Based on their saliency characteristics and the target level, the proposed algorithm decides the maximum coding tree depth for each coding unit. The certain saliency threshold is used to decide whether the coding unit should be split into smaller CUs. By this way, the coding complexity control can be achieved with minimal loss of visual distortion.