【作者】 张智澄；

【导师】 郭炜；

【作者基本信息】 上海交通大学 ， 电路与系统， 2007， 硕士

【摘要】 目前大多数的视频标准例如H.264、MPEG-4、AVS都是采用基于分块的离散余弦变换(BDCT)的压缩编码方式。这是一种将视频图像分成块并独立处理的方法,其最大的缺点就是在比特率不太高的情况下会出现方块效应,因此,在视频系统中去块滤波技术就显得十分重要。去块滤波技术主要分为环路滤波和后处理滤波。H.264和AVS都定义了环路去块滤波算法,可以显著地提高运动估计的质量,但是其并没有太多地考虑到主观视觉效果,因此,即使在编解码系统中包含环路去块滤波器,还是需要在接收端添加后处理去块滤波器以提高最终的主观视觉效果。另一方面,和环路滤波相比,后处理去块滤波更具有广泛性和通用性。本课题的主要研究方向是视频去块滤波算法的研究与实现。首先介绍了方块效应产生的原因、去块滤波器的分类和人类视觉系统。接着介绍了当前主流的环路去块滤波算法和后处理去块滤波算法。然后,根据人类视觉系统的特点,对现有后处理去块滤波算法中的模式、参数以及区域判别方法作了系统的研究及大量的实验,以改善主观视觉效果及降低实现复杂度为目的,对现有算法进行了一定的改进。利用Matlab对滤波结果进行考察,结果表明,本文所提出的改进算法同样具有较好的滤波效果,并且提供了降低设计方案实现成本的理论基础。之后,本文对改进后的算法进行了RTL级硬件实现并使用0.18um TSMC的工艺库进行综合。结果显示,该算法和硬件结构能够在使用较少存储单元的情况下,仍然达到视频实时处理的速度要求。最后,针对硬件加速器在灵活性上的不足,运用MOVE处理器,利用其配置性强和扩展性高的特点,在确保基本达到性能指标的前提下实现了算法的灵活移植。更多还原

【Abstract】 Most of the current video encoding and decoding compression standards such as H.264, MPEG-4 and AVS are based on BDCT (Blocked Discrete Cosine Transformation). However, this is a method that dividing the image into blocks and independently processing each block. Its biggest disadvantage is that it will cause blocking effect in low bit rate. Thus it is necessary to include deblocking filter in the whole video system.Deblocking filter can be divided into two types: in-loop filter and post filter. H.264 and AVS have both defined the in-loop deblocking algorithm to increase the quality of motion estimate. However, it doesn’t concern much about subjective visual effect. Thus post deblocking filter should be added to video systems to increase subjective visual effect even if in-loop filter is included. On the other hand, compared with in-loop filter, post filter is more general and has been widely used. This thesis mainly focuses on the research and implementation of deblocking algorithm in video processing system. The paper first introduces the generation of blocking effect, the classification of deblocking filter and human visual system. Then in-loop and post filter is introduced in detail.Based on the feature of human visual system, the paper does research and many experiences on the mode, parameter and mode decision method of current post deblocking algorithm to make some improvements. All these are aimed at improving subjective visual effect and decreasing complexity in realization. Using Matlab to observe the result and it shows that the improved algorithm also has good effect while providing the academic basis of decreasing the cost of implementation.Based on the research, the proposed algorithm is implemented in the paper by hardware accelerator in RTL level and synthesizes it using 0.18um TSMC technical library. It is shown in the results that the proposed algorithm and hardware architecture can meet the real-time processing requirement using less memory. Since the hardware accelerator is not flexible, the paper using MOVE processor to realize the algorithm and making advantage of its high compatibility and strong scalability. By doing this, its flexibility is improved while ensuring the speed and performance.更多还原