视频编码中自适应环路滤波器的算法与硬件架构.pdfVIP

Abstract Abstract As one of the most promising tools, Adaptive Loop Filter (ALF) is introduced into HEVC coding structure to obtain better coding efficiency. However, ALF increases the burden on computational complexity significantly due to utilize of wiener filter, which limits its application. In this thesis, we focus on reducing ALF complexity while maintain its high coding efficiency. Firstly, this thesis proposes an adaptive subsampling method based on local direction feature. This proposal explores the noise property and subsampling method. Subsampling will worsen aliasing problem in high frequency region. Therefore, the subsampling pattern can be decided according to the frequency property in local region. By reusing directional feature in ALF, proposed method can indicate the local frequency property effectively while not introduce extra computational burden. With proposed directional feature based subsampling algorithm, more than 38% ALF processing time can be saved while the BDRATE only increases 0.1% on average. Based on the adaptive subsampling method, a four pixel parallel hardware architecture is proposed, which can deal with 1920 ×1080@60fps video sequence. Proposed architecture contains register array, calculation unit, parallel to serial convertor, memory and controller. Taking advantage of ALF symmetric property, proposed architecture adds pixel values at symmetric position and then calculates correlation matrix. Therefore, number of multiplication can be reduced. Combining with proposed subsumpling algorithm, it requires only 2 pixels for each 2 ×2 block at the worst case. In order t