【作者】 杨洋；

【导师】 沈礼权；

【作者基本信息】 上海大学 ， 通信与信息系统， 2019， 硕士

【摘要】 多媒体与计算机技术的发展,促成了动画片、远程教育视频等计算机生成的屏幕内容视频的普及。为了提高屏幕内容视频编码的性能,专家学者以高效视频编码标准(High Efficiency Video Coding,HEVC)为基础,提出了HEVC的屏幕内容编码扩展标准(HEVC-Screen Content Coding,HEVC-SCC),但HEVC-SCC中的码率控制部分依然沿用了HEVC中的R-λ模型。R-λ模型是根据传统的自然视频编码后的失真与编码比特之间的关系提出的,而屏幕内容视频在编码后这两者之间的关系发生了变化,R-λ模型不能准确反映屏幕内容视频编码失真和比特之间的关系,并且该模型的比特分配和参数更新方法也不能兼顾屏幕内容视频的特性。本文对屏幕内容视频的特性进行了研究,根据屏幕内容视频的特性提出了有针对性的码率控制算法,主要研究内容如下:1.基于屏幕内容视频内容的R-λ模型优化及帧内比特分配算法。由计算机生成的屏幕内容视频中包含文字,内容简单、纹理平坦的屏幕图像以及自然图像,而这些不同的视频内容在编码后产生的失真和编码比特之间的关系也是不一样。基于这样的发现,我们首先将屏幕内容视频中的编码树单元(Coding Tree Unit,CTU)划分为三类:文字-CTU,屏幕图像-CTU和自然图像-CTU。接着为这三类CTU分别建立R-λ模型,并依据各自的模型分别进行参数更新。然后在CTU级进行比特分配之前,先进行区域级的比特分配。区域级的比特分配是指分别为仅包含某一种CTU的区域分配比特。最后在区域级的目标比特的基础上对每一个CTU进行比特分配。实验结果表明,相较于HEVC-SCC中的默认码率控制算法,在Random Access的编码结构下,本算法的BDPSNR提高了0.637d B,而在Low Delay的编码结构下,本算法的BDPSNR提高了0.749d B,实现了良好的码率控制性能。2.基于屏幕内容视频特性的GOP级及帧间比特分配算法。相较于传统的通过摄像机拍摄的自然视频,计算机生成的屏幕内容视频场景切换频繁,基于这一特性,本文提出了针对于屏幕内容视频的比特分配算法。首先,根据视频帧间的均方误差(Mean Square Error,MSE)将屏幕内容视频中的所有帧分为场景切换帧和场景未切换帧。依据图像组(Group of Pictues,GOP)中的场景切换帧的数目进行GOP级的比特分配。在帧级的比特分配中,场景切换帧参考的是帧间的MSE的比值,场景未切换帧参考的是帧内、帧间复杂度与编码比特之间的关系。实验结果表明,相较于HEVC-SCC中的帧级平均分配比特的算法,本文提出的算法的BDPSNR提高了1.874d B,而相较于HEVC-SCC中的帧级非平均分配比特的算法,本文提出的算法的BDPSNR提高了0.875d B。这样的实验结果证明了我们提出的算法的优越性。更多还原

【Abstract】 As the development of multimedia and computer,a new type of video named as screen content video has becoming more and more popular.This type of video is generated by the computer.For example,animations and remote education videos are typical screen content videos.In order to improve the coding efficiency of screen content videos,some specialists extend the High Efficiency Video Coding(HEVC)standard and propose a new video coding standard to encode screen content videos,and name it as HEVC Screen Content Coding extension(HEVC-SCC).However,the rate control algorithm in HEVC-SCC is inherited from HEVC and the inherited R-λ rate control model is designed for natural content videos.The establishment of R-λ model is based on the relationship between the coding bits and distortion(R-D).But in fact,the R-D relationship of screen content videos is changed.Moreover,this R-λ model does not take the characteristic of screen content video into consideration when the bit allocation and parameter update operations are conducted.Therefore,the R-λ model is not suitable for screen content videos.This paper takes several researches on the characteristics of screen content videos,and proposes two rate control algorithms for screen content videos.The major contents of the dissertation are summarized as follows:1.The optimization of R-λ model and intra-frame bit allocation method based on video contents for the screen content video.Contents in the screen content video contain text,screen images with low texture complexity and nature images captured by cameras.Coding these different video contents results in different R-D relationships.In view of this characteristic,the rate control method should be conducted for different video contents separately.Firstly,the coding tree units(CTU)in screen content videos are claasified into three types: Text-CTU,Screen ImageCTU and Nature Image-CTU.Then,three R-λ models reflecting different R-D relationships of different CTUs are established,and the model parameters are updated on the basis of their own models.Furthermore,a bit allocation scheme for the region containing only one kind of CTU is determined before allocating bits for each CTU.Experimental results show that the proposed rate control algorithm achieves 0.637 d B BDPSNR increase under the Random Access configuration and 0.749 d B BDPSNR increase under the Low Delay configuration on average,compared to the default R-λ models in HEVC-SCC reference software.2.A GOP and frame level bit allocation algorithm based on the characteristics of screen content videos.Compared with the conventional nature video which is captured by cameras,the scene in screen content videos changes frequently.The bit allocation algorithm in this paper is proposed according to this characteristic.Firstly,the mean square error(MSE)of adjacent frames is used to classify all frames in screen content videos into two types: scene changed frames and scene unchanged frames.At the GOP level,the bit allocation scheme is based on the number of the scene changed frames in the current GOP.At the frame level,the bit allocation scheme for scene changed frames is according to the ratios of MSE of all frames in a GOP,and the bit allocation scheme for scene unchanged frames is according to the relationship between the inter and intra frame complexity and coding bits.Experimental results show that the proposed bit allocation algorithm achieves 1.874 d B BDPSNR increase,compared to the default R-λ model with an equal bit allocation method at frame level in HEVC-SCC.In comparison with the default R-λ model with an unequal bit allocation method at frame level in HEVC-SCC reference software,the proposed algorithm achieves 0.875 d B BDPSNR increase.These experimental results verify the efficiency of our proposed algorithm.更多还原