Fast Intra Encoding Decisions Based on Horizontal-Vertical Progressive Gradient Accumulation for HEVC

High Efficiency Video Coding (HEVC) employs a flexible quad-tree structure for intra prediction. It enables up to 35 prediction modes with using all the possible depth levels and prediction modes to obtain the optimal CU (coding unit) and PU (prediction unit) mode. However, these characteristics bring great computational complexity. To address this issue, we propose a progressive gradient accumulation strategy for intra encoding decisions. The horizontal and vertical progressive gradient values will be obtained by adding up all the differences of every adjacent column and row. Then, the maximum optimal depth of the neighboring CU with similar vector is used for current CU early termination, and the vector direction is used for current PU mode decision. In the process of arithmetic, the accumulative values will be simplified into only calculating the difference of edge pixels. Experimental results verify that the proposed algorithm can reduce 21.82% coding time with 3.89% BDBR increasing.