HEVC环内滤波算法优化

发布时间：2018-05-01 08:17

本文选题：HEVC + 环内滤波算法　；参考：《西安电子科技大学》2014年硕士论文

【摘要】：随着人们对视频更高质量不断追求,高清或超高清视频处理已成为现代视频技术的发展趋势。然而,高清视频的大量应用给视频传输与存储带来了巨大的挑战。为获得更高的压缩效率,最新一代视频编码标准HEVC延续了H.264标准中混合编码框架与核心思想,同时增加了很多新的视频压缩技术,例如环内滤波算法相比于H.264有了很大改进。环内滤波算法旨在消除编码过程中预测、变换和量化等环节产生的失真,为运动补偿预测提供更高质量的参考帧,它主要包括去方块滤波算法和样值自适应偏置算法。本文通过分析HEVC环内滤波算法的基本原理,并在HM12.0软件环境中对参考视频序列进行测试对比,发现针对4:4:4格式和位深度大于10比特的两种视频序列,HEVC通常均无法获得理想的效果,而且存在编码效率下降的问题。基于这一问题,本文分别设计4:4:4格式色度去方块滤波优化算法和高位深度SAO偏移值自适应伸缩算法,可以有效地提升视频主观质量和编码效率。1.在4:4:4格式色度去方块滤波优化算法中,为提升4:4:4格式色度去方块滤波的性能,本文分别设计了两种算法来进行色度分量去方块滤波优化。第一种是基于滤波强度去方块滤波算法,该算法是将亮度去方块滤波运用于色度分量;第二种是基于滤波范围色度去方块滤波算法,该算法则是根据合适的滤波范围,启动色度滤波算法。上述两种算法都是充分利用色度分量所包含的信息进行滤波,并为色度分量增加独立的去方块滤波控制参数。进一步,利用JCT-VC通用的测试条件对4:4:4格式色度去方块滤波优化算法进行验证,实验结果表明在RA和LB Main-tier下编码效率得到了提升:第一种算法提升了0.2/2.3/2.2和0.4/2.1/1.9(Y/U/V BD-rate(%)),第二种算法提升了0.1/0.5/0.6和0.0/0.4/0.5。2.在SAO偏移值自适应伸缩算法中,继续采用SAO技术思想,即在LCU内部将重建像素划分为带状偏移(Band Offset)和边缘偏移(Edge Offset)类型,并为它们加上相应的偏移值来减少各个区域像素失真,不同的是,本算法通过使用合适的伸缩因子对SAO偏移值进行自适应伸缩变换,其核心思想是分段线性变换。该方法处理灵活,并且易于硬件实现,当位深度大于10比特时,SAO偏移值在较低幅值时有更精确的伸缩,同时兼顾较大幅值,从而提高了编码滤波效率。实验结果表明,当在All-Intra-HE-Main-Tier,IBDI=2和IBDI=4时,编码效率分别提升了0.3/0.2/0.2(Y/U/V BD-rate(%)),0.3/0.4/0.4(Y/U/V BD-rate(%))。采用该自适应伸缩算法能够更加灵活地适应不同特征输入序列。
[Abstract]:With the continuous pursuit of higher video quality, HD or UHD video processing has become the development trend of modern video technology. However, the application of HD video brings great challenges to video transmission and storage. In order to achieve higher compression efficiency, the latest generation video coding standard HEVC extends the hybrid coding framework and core idea of H.264 standard, and adds a lot of new video compression techniques, such as the improvement of the filtering algorithm in the loop compared with H. 264. In order to eliminate the distortion caused by prediction, transformation and quantization in the coding process, the intra-loop filtering algorithm provides a better reference frame for motion compensation prediction. It mainly includes the de-block filtering algorithm and the sample value adaptive offset algorithm. In this paper, the basic principle of HEVC filter algorithm is analyzed, and the reference video sequences are tested and compared in the HM12.0 software environment. It is found that the two kinds of video sequences with 4:4:4 format and bit depth greater than 10 bits can not get ideal results, and the coding efficiency is reduced. Based on this problem, this paper designs the 4:4:4 format chroma de-block filtering optimization algorithm and the high depth SAO offset adaptive scaling algorithm, which can effectively improve the subjective quality and coding efficiency of video. In order to improve the performance of 4:4:4 format chroma de-block filter, two algorithms are designed to optimize the chrominance component de-block filtering in 4:4:4 format. The first is based on the filter intensity debonding algorithm, which applies the luminance de-block filter to the chrominance component, and the second is based on the filtering range chroma de-block filtering algorithm, which is based on the appropriate filtering range. Start chrominance filtering algorithm. Both of the above algorithms make full use of the information contained in the chrominance component to filter and add independent de-block filter control parameters to the chrominance component. Furthermore, the 4:4:4 chrominance de-block filtering optimization algorithm is verified by using the general test conditions of JCT-VC. The experimental results show that the coding efficiency is improved in RA and LB Main-tier: the first algorithm increases 0.2 / 2.3 / 2.2 and 0.4/2.1/1.9(Y/U/V BD-rate2, and the second improves 0.1 / 0.5 / 0.6 and 0.00.4 / 0.5.2. In the adaptive scaling algorithm of SAO offset, the idea of SAO technology is adopted, that is, the reconstructed pixels are divided into banded offset sets and edge offset edge offsets in LCU, and the corresponding offset values are added to them to reduce the pixel distortion in each region. The main idea of this algorithm is piecewise linear transformation. The method is flexible in processing and easy to be implemented in hardware. When the bit depth is more than 10 bits, the Sao offset value has a more accurate scaling when the amplitude is lower, and at the same time, it takes into account the larger amplitude, thus improving the efficiency of coding filtering. The experimental results show that at All-Intra-HE-Main-Tiern IBDI2 and IBDI2 = 4, the coding efficiency is improved by 0.3 / 0.4 / 0.4 / 0.4Y / U / V BD-rate2, respectively. The adaptive scaling algorithm can adapt to different feature input sequences more flexibly.
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN919.81;TN713

【参考文献】