【作者】 吴婧；

【导师】 顾伟康；

【作者基本信息】 浙江大学 ， 通信与信息系统， 2009， 博士

【摘要】 随着图形化用户界面GuI(Graphic user Interface)的广泛应用和计算机网络的迅速发展,实时的桌面图像序列传输技术正得到越来越广泛的关注,在网络多媒体会议、产品演示、远程办公、远程教学等领域都具有非常广阔的应用前景。利用这种技术,不但可以在当前的数字设备上,如PC机或便携式显示设备,与远程的其它计算机进行信息交互,而且还可以操控远程的计算机来解决当前资源无法解决的问题,实现广泛的资源(包括软件资源和硬件资源)共享。但是由于桌面图像序列的原始数据量巨大,在Internet上传输极易引起网络拥塞和传输延时,所以实时传输前必须对桌面图像序列进行有效的压缩编码。桌面图像序列包含了文本、图形和自然图像的信息,是一种混合图像序列。如何有效地区分桌面图像序列中的文本、图形和自然图像信息,并根据各自的特性进行压缩编码,是提高桌面图像序列压缩效率的关键。本文针对桌面图像序列编码中基于区域特征的分类编码、分级编码和编码效应后处理3个方面进行了深入的研究。本文首先分析了桌面图像序列的特性,在此基础上,提出了一种新的基于“块”区域特征的桌面图像序列编码方法(BRC-SISC,Block Region Character based Screen ImageSequence Coding)。BRC-SISC算法使用帧间运动检测去除桌面图像序列时域上的冗余信息,并基于颜色和梯度特征的不同表现,区分出文本/图形块、图像块和混合块。根据块分类的结果和各自的特性,文本/图形块采用调色板编码、Hextile编码和Deflate编码相结合的无损压缩方式(PHDC,Palette,Hextile and Deflate Combined coding),图像块采用Jpeg有损压缩编码。混合块则在精细分类的基础上,引入“层”的概念,填充为文本/图形“层”块和图像“层”块分别编码。算法能够准确地区分出文本/图形区域和自然图像区域,有效地减少或避免了振铃效应的产生。同时该算法的压缩效率较高,运算速度较快,满足实时系统的要求。接着本文针对桌面图像序列传输系统中,多点接入的客户端在用户要求、终端计算能力和网络带宽上存在着差异的问题,提出了一种基于“块”区域特征的桌面图像序列分级编码方法(BRC-SSISC,Block Region Character based Scalable Screen Image SequenceCoding)来满足不同性能客户端的需求。其基于前面文本/图形信息和图像信息的分类结果,依据人眼视觉的敏感特性,提供了3层的SNR可分级性,最大限度地为所有的客户端提供了优质的桌面图像码流。同时本文还提出了一种补偿编码机制(CC,CompensatoryCoding),以确保在客户端性能提高或桌面图像序列变化较少的时候,为客户端补偿提供未变化块的更清晰的码流。实验结果表明,BRC-SSISC算法能够有效地提供分级压缩的图像质量,且算法的复杂度较低,运算速度较快,满足实时传输的需求。最后本文讨论了重建图像中编码效应的后处理方法。通过对块效应和振铃效应的形成原因和表现形式的分析,发现受图像局部区域特征和人类视觉特性的影响,编码效应在不同的图像区域会有不同的表现和视觉感受。因此,本文提出了一种新的空域自适应的编码效应消除算法(SAFCAR,Spatial Adaptive Filtering of Coding Artifacts Removal)。块效应滤波时,SAFCAR算法依据纹理信息将图像块边界区域划分为平滑区域和纹理区域,分别采用不同的检测标准和滤波范围来消除块效应。振铃效应滤波时,SAFCAR算法采用了边缘的检测进行强振铃效应跟踪,并增加了补充检测来对弱边缘区域和编码效应扩散区域的振铃效应强度进行判断。检测完成之后,各区域根据振铃效应的强度也采用了不同强度的平滑滤波。SAFCAR算法能够有效地消除自然图像序列和桌面图像序列中的块效应和振铃效应,并较好地保护图像的纹理细节信息。更多还原

【Abstract】 With the wide spread application of GUI(Graphic User Interface) and the rapid development of networks,Real-time screen image sequence transmission technology become more and more important,and will be widely used in the fields of internet multimedia conference,products demo,telecommuting,remote education etc.Due to this technology, people can use local digital devices like PC and smart display set to interact with remote computers graphically,or to realize remote computer aided massive computing work.This makes a wide spread hardware and software resource share throughout networks.However,the large amount of original image data is a major obstacle for real-time screen image sequence transmission.It usually causes network congestion and transmission delay over Internet. Therefore,screen image sequence compression algorithms are essential for real-time transmission.Screen image sequence is a kind of compound image sequence,containing text,graphics and natural images.How to compress screen image sequence is a hard problem.An effective region classification and a proper hybrid encoding are very helpful to improve its compression efficiency.This dissertation focus on the screen image sequence coding and researches the methods of region-based classification and encoding,scalable encoding and coding artifacts removal.Firstly,this dissertation proposes a new block region character based screen image sequence coding scheme(BRC-SISC).It applies the inter-frame motion detection to eliminate the temporal redundant information.And according to the different characters on color and gradient,it classifies each block of image into three categories:text/graphics block,picture block and hybrid block.Text/graphics block is compressed with a lossless coding algorithm (PHDC),which integrates palette coding,Hextile coding and Deflate algorithm.Picture block is encoded by lossy JPEG.For hybrid block,first,a second classification is adopted to distinguish between text/graphics pixels and picture pixels.Then,each kind of pixels is filled into a layer block.Text/graphics layer block uses lossless coding method,and Picture layer block adopts lossy coding method.The proposed classification and coding methods could exactly distinguish between the text/graphics and picture information.And it achieves good compression ratio with low computation,which satisfies the real-time application requirement.Secondly,differences in consumer requirement,terminal processing capability and network bandwidth always exist among clients in screen image sequence transmission system. In order to meet the requirements of the clients in different performance conditions of computer and network,this dissertation proposes a block region character based scalable screen image sequence coding method(BRC-SSISC).First,based on the earlier text/graphics and picture classification result and the human vision character,a three-layer SNR scalable coding method is offered to provide high quality screen images for all the clients.Then,a compensatory coding method(CC) is applied to make sure clearer screen image could be received,when the performance of client becomes better or screen image sequence changes slower.Experiment results shows that the proposed scalable coding method is effective.And it also achieves good compression ratio with low computation,which satisfies the real-time application requirement.Finally,coding artifacts removal method is discussed.Influenced by the local region character of image and the human vision character,coding artifacts are manifested to varying degree in different areas.Therefore,this dissertation proposes a new spatial adaptive filtering of coding artifacts removal(SAFCAR).For de-blocking,it classifies the block boundary region into smooth region and texture region based on its gradient character,and then adopts different detecting standards and filtering range to remove the blocking artifacts.For de-ringing,besides the edge detection used to trace the strong ringing artifacts block,a complementary ringing detection method is proposed to locate the weak ringing artifacts block.After twice detections,a fuzzy filter,which is adaptive to the ringing strength,is applied to remove ringing artifacts.In both natural picture sequence and screen image sequence,the proposed post-filtering method has a good performance on coding artifacts removal and detail preservation.更多还原