Expediting Determination of Coding Unit Partition Based on Quantization Parameter and Frame Complexity

Abstract

Choosing current major video coding standard, called H.264, to compress ultra-high resolution videos will cause low performance of compression in bitrate because of fixed macroblocks in H.264. For solving the large bitrate problem, a new video coding standard called High Efficiency Video Coding (HEVC) includes a quadtree-based Coding Unit (CU) block partitioning structure is established. However, HEVC causes a significant increase in computational complexity, compared to H.264 due to the necessity of searching the optimal Largest Coding Unit (LCU) partition. In this paper, we propose the fast luma-based CU partition (FLCUP) for HEVC intra prediction coding, which reduces significant computational complexity with negligible distortion in rate-distortion performance. FLCUP performs two steps: 1) analyzing the complexity of an original frame and 2) predicting LCU partition. Current the complexity of an original frame of LCU is analyzed by Discrete Wavelet Transform and several thresholds are obtained from experimental results. After analyzing the complexity of an original frame, FLCUP could predict the LCU partition for encoder without the unnecessary depth searching. Experimental results show that compared to the standard HEVC, FLCUP reduces the computational complexity by 38.63% at the cost of increasing only 1.62% in BD rate.