基於畫面間運動向量之HEVC運算複雜度分配與控制

摘要

High Efficiency Video Coding (HEVC)是一種新一代的視訊壓縮技術,近年來由JCT-VC團隊所開發與制定,相較於過去的視訊壓縮標準H.264/AVC,HEVC提供更多新壓縮技術來達到更高的編碼效率,其適用範圍從Ultra HD (4000x2000或更高)到行動裝置上HD (720p或1080p) 的影片。在眾多新技術中,HEVC提供可調式的編碼單位(coding unit, CU),其預設大小從8x8到64x64不等,取代H.264中固定大小的編碼單位(MarcoBlock, MB),雖然可以大幅減少“位元率 (bitrate)”,但也相對地大幅增加計算複雜度。然而在行動裝置上受限於電源容量,可用的運算複雜度通常是有限的,因此,降低並控制視訊編碼器的運算複雜度以延長行動裝置的使用時間,並且維持較佳的視訊位元率-失真效能是非常重要的議題。 本論文針對HEVC高解析度視訊於運算資源有限的行動裝置上,提出一套編碼運算複雜度分配與控制的機制,包括視訊層到畫面層、最大編碼單位層之分配與控制,在每一層的分配上都採取不同的分配方式,以達到最好的編碼效率為目的。首先在畫面層級我們根據每張畫面的量化參數(QP)給予其不同的運算複雜度,接下來利用前一張畫面所得的運動向量(MV)的資訊,將運算複雜度分配置每個最大編碼單位(Largest CU, LCU)層級,並且決定是否要切到下一深度。本論文所提出的方法在總體複雜度節省40%的情況下可達到BD-bitrate僅提升2%,BD-PSNR降低不到0.1 dB,而控制誤差僅有0.4%。

 

Motion Vector-based Computational Complexity Allocation and Control for HEVC Inter Coding

Abstract

The latest video compression standard HEVC provided the coding unit (CU), defined by quad-tree structures, to achieve high coding efficiency. Compared with previous standards, HEVC encoder increases much computational complexity. However, the allowable computational capability of a portable device for real-time video encoding is generally constrained. Therefore, complexity control of HEVC and maintain the optimal rate-distortion performance is important.

We proposed a method of constrain the complexity for high resolution video on mobile device, including computational complexity allocation and control from video layer to frame layer, largest coding unit (LCU) layer. In order to reach the best performance, the method of allocation in each layer, we take difference method to allocate the complexity to each layer. In frame layer, we allocate complexity based on the quantization parameter (QP) of each frame. Based on the information in previous frame, we allocated the complexity to LCU layer, and decide whether split to next depth. The experimental results show that under 60% target computational complexity, the loss of average BD-PSNR and BD-bitrate is negligible and the complexity control error is no more than 0.4%.

 

 

Index terms – HEVC, complexity control, complexity allocation, coding unit