【作者】 丁志鸿；

【导师】 刘立柱；

【作者基本信息】 解放军信息工程大学 ， 军事情报学， 2011， 博士

【摘要】 随着数据压缩技术和通信技术的发展,多媒体应用迅猛发展。数据压缩技术一方面减少了图像视频数据的传输量,另一方面由于压缩数据间相关性的降低而加重了传输过程中误码对图像视频恢复质量的影响。实际通信中,由于环境和设备等因素影响,误码的产生、数据的丢失难以避免,由于压缩码流极易发生误码扩散,会导致图像视频质量严重下降,甚至是无法恢复。现有的抗误码技术如纠错编码和重传机制等,各有其优缺点和适用环境。其中信源解码阶段的抗误码技术不增加编码冗余,与标准兼容,且与信道无关,尤其是在某些特殊应用环境中,信源解码抗误码技术是唯一的选择。论文针对图像和视频压缩数据的特点,分别研究了图像、视频信源码的纠错译码技术和错误隐藏技术。在分析两类抗误码技术优缺点的基础上,研究了信源解码阶段的抗误码新机制。论文的主要贡献体现在如下六个方面:(1)研究了信源解码抗误码新机制。融合接收端抗误码技术,将智能处理技术与译码技术相结合,在误码检测、纠错译码技术研究和错误隐藏技术研究中,采用了模式识别等智能技术,可以充分发挥纠错译码技术、尽早同步技术和错误隐藏技术各自的优势,有效解决误码定位等难题。(2)研究了静止图像信源解码阶段的纠错译码技术,分别提出了基于概率统计的传真MMR码纠错译码算法和针对高权重比特域突发错误的JPEG2000纠错译码算法。前者通过大量的MMR码样本,训练得出含错行比特数以及等长序列的概率模型,纠错前使用基于贝叶斯理论的最大后验估计对模型进行动态更新,由此指导纠错译码,纠错速度较以往算法有了明显的提高。后者针对JPEG2000数据区不同比特对图像恢复质量影响程度的不同,在检测到误码之后,在高权重比特域,使用滑动窗口对误码进行搜索和定位,获得了较好的效果。(3)研究了视频码流错误检测技术,提出了一种基于多特征的视频分级错误检测算法。在研究现有视频码流错误检测技术的基础上,根据模式识别理论,总结了一系列常用的检错特征,同时引入了新特征,然后使用了基于语法、基于阈值和基于支持向量机的三级判决准则进行错误检测。实验表明,该算法可以检测到视觉表现不明显的错误块,提高了错误块检测率。(4)研究了视频空域错误隐藏算法,提出了一种基于边界分段匹配的空域视频错误隐藏算法。根据计算机视觉理论,先对边缘进行估计,根据边缘点的分布及方向对边界进行分割,使用各分段在邻近区域内进行匹配填补,经过迭代各边界按一定策略向内填补最终完成错误隐藏。实验表明,本算法有效地保留了边缘信息,使恢复图像的主客观质量有较大的提高。(5)研究了视频时域错误隐藏算法,针对连续宏块丢失的情况,提出了一种适用于H.264的基于运动物体的时域错误隐藏算法。根据视觉注意机制,首先分析丢失宏块相邻宏块的运动矢量及灰度值分布,然后对平坦区域及灰度一致的运动区域先进行错误隐藏,最后对遗漏像素进行隐藏。实验表明,该算法能很好地保持物体形状,减少块效应,提高了恢复质量。针对整帧丢失的情况,提出了一种基于残差分布的H.264自适应整帧恢复算法。根据视觉注意机制,首先对参考帧的残差信息进行分析,对于平坦区域或刚性运动区域,采用传统的运动矢量复制法,而对运动复杂区域或产生形变等区域进行基于像素的运动矢量重新估计,再以光流法进行恢复。实验表明,本算法在主观视觉质量和客观峰值信噪比上都优于传统方法。(6)在信源解码抗误码新机制的框架下,分别设计了传真MH码二值图像和MPEG-2视频图像抗误码融合技术。前者在译码完成后,对错误扫描线分布进行统计。若错误扫描线为孤立扫描线,则使用新误码补偿技术处理当前扫描线;否则使用滑动窗口在当前行数据流内进行纠错译码;若纠错失败,再使用新误码补偿技术。后者在检测到错误宏块时,判断图像块是否为平坦块,若是则使用错误隐藏技术恢复错误宏块;否则从该宏块编码数据的开始位置尝试纠错译码;若纠错失败再依次使用尽早同步技术和错误隐藏技术。实验表明,信源码抗误码新机制较好地缓解了提高抗误码性能与运算速度之间的矛盾。目前,以本论文研究为基础的系统已在相关单位实际工作中得到使用,发挥着重要作用。更多还原

【Abstract】 With the development of data compression and communication technology, multimedia applications are growing rapidly. On the one hand, data compression technologies make less data that needs to be transmitted, on the other hand, as the correlation between the compression data is reduced, the effect of error to image and video quality increases. During the actual signal transmission, many reasons such as entironment and equipment make the transmission error and packet loss unavoidable. And the errors are prone to spread, resulting in a serious decline in the quality of image and video, even the failure of recovery The existing error resilience technologies such as error-correcting codes and retransmission mechanism, have their own advantages and disadvantages. Each is suitable for different environment. Among these technologies, error resilience technologys in source decoder does not increase the code word redundancy, and is standard compliant, and independent on transmission channel. Especially in some special applications, the error resilience technology in source decoder is the only choice.According to the characteristics of image and video compressed data, error correction and error concealment technologies of image and video are researched in this thesis. A new mechanism based on the analysis of the error resilient technologies in source decoder is researched. The main contribution of the thesis includes six aspects as follows:(1) A new mechanism for error resilience in source decoder is researched. It is the integrated application of the error resilience technologies in source decoder, and combines intelligent processing technologies with decoding technologies. In error detection, error correction and error concealment, intelligent technologies such as pattern recognition are used. The mechanism can give full play to error correction technology, early resynchronization technology and error concealment technology to their own advantages and effectively solve many difficult problems such as error localization.(2) Error correction decoding technologies to the still image are researched. Fax MMR error correction decoding algorithm based on probability statistics and JPEG2000 error correction decoding algorithm against the high weighted bits are proposed respectively. Using a large number of MMR code samples, the former algorithm gets the number of error line’s bits and the probability of the same length of different sequences. The probability model is updated dynamically by maximum posteriori estimation based on Bayes theory before error correction. The results are used to guide the error correction decoding. As a result, the speed of the propsed algorithm is faster than the past one. Because different bits of the JPEG2000 have different influence to image restoration quality, once some bit errors are detected, a sliding window is used to search errors in the high weighted bits region. The experiment shows its good performance.(3) Video stream error detection technologies are researched. A hierarchical video error detection algorithm is proposed based on mutifeatures of macroblock. The existing video streams error detection technologies are studied firstly. Secondly according to the pattern recognition theory, a series of error detection features which are commonly used are summrized and some new characteristics are introduced. And finally the hierarchical error detection algorithm based on the syntax-based method, threshold method and support vector machine method is used to detect errors. Experiments show that the algorithm can detect errors in blocks which visual expression is not obvious and improve detection performance for error block.(4) Video spatial error concealment algorithms are researched. A new spatial error concealment method based on match of boundary subsection is proposed. According to computer vision theory, it first estimates edge of each block boundary. Then according to distribution and direction of the edge points, the boundaries are partitioned into several subsections. Pixels of each subsection are recovered by the best match pixels in the vicinity of the subsection. Concealment is accomplished by reconstructing each boundary repeatedly according to a suitable strategy. Experimental results show that the new algorithm can improve the subjective and objective quality of recovering image as it can effectively retain the edge information.(5) Video temporal error concealment algorithms are researched. For the case of consecutive macroblocks loss, a new temporal error concealment method for H.264 based on motion object is proposed to reconstruct the missing blocks. According to the visual attention mechanism, the method analyses the distribution of the motion vectors and the gray for adjacent macroblocks of the missing macroblocks firstly. Secondly, according to it, the smooth region and the motion region with similar gray values are concealed first. Finaly, the missing pixels which still exist are concealed. Experimental results shows that the new algorithm can keep the shape of motion object better and decreases the block effect. For the case of the whole frame loss, an error concealment algorithm based on residual distribution is proposed for whole frame packet lost in H.264. According to the visual attention mechanism, residual information of reference frame is analyzed. According to the result, motion vector copy algorithm is used for the region where the image is smooth or the image object’s motion is rigid. To the other region, the motion vectors of each pixel are revaluation, and then optical flow algorithm is introduced. Experimental results show that the algorithm has better performance than traditional algorithms on both visual effect and PSNR.(6) According to the framework of the new mechanism for error resilience, fusion technologies of MH code and MPEG-2 code are proposed. For MH code, the distribution of the wrong scan lines is analysed after decoding. If the error scan line scan is isolate, a new error compensation technology is used to deal with the current scan line. Otherwise a error correction technology is used to recover the errors in the cunrrent scan line. If error correction fails, the compensation technology is used finally. For MPEG-2, after some error MBs are detected, whether the lost MB is flat or not is judged firstly. If it is a flat MB, error concealment techonoligies are used. Otherwise, error correction technology is used. If error correction fails and then some early resynchronization technologies and error concealment technologies are used in order finally. Experiment results show that the new mechanism can mitigate the conflict between the speed of calculation and the high performance of error resilience in a certain extent.At present, this research-based system has already been used in the relevant units and plays an important role in practical.更多还原