压缩感知视频传输中质量评价及相关问题研究

发布时间：2018-06-19 09:54

本文选题：压缩感知 + 压缩感知视频　；参考：《南京邮电大学》2016年博士论文

【摘要】：在现今信息技术大发展时代,以图像视频为载体的服务应用在信息传输中占据重要地位,与此同时图像视频处理传输技术得到了普遍关注。以H.264为代表的成熟且高效的预测类视频编码方法是图像视频处理的基础,但是,由于预测类编码所需计算量巨大,并且会导致信道误差影响延续,不适用于一些资源有限、低耗要求的无线应用场景,如参与式感知、无线多媒体传感器网络等。压缩感知(CS)理论为解决上述问题提供了一种新的思路。该理论不仅在采样速率方面突破了奈奎斯特频率的限制,而且在采样观测的同时以降维的方式实现了图像或视频数据一定程度上的压缩,实现了采样和压缩两个过程的合并,避免了传统先采样后压缩所产生的资源耗费。而且,压缩感知作为一种用信息全局观测(即所谓信息采样)代替传统的信号局部采样(即所谓信号采样)的新型信号采集方式,通过线性随机投影所获得的观测值承载了具有相同重要性的信号整体信息,有利于构造简单有效的抗信道误差方案。压缩感知理论与现有视频技术结合,可有效缓解高速采样实现的压力,减少节点采集、编码、处理和传输的成本,将会进一步推动信息领域的向前发展。本论文在压缩感知理论的基础上,以图象/视频信号为主要研究对象,紧紧围绕视频质量评价、编码效率提高和图像/视频差错控制三个方面中与视频传输质量有关的关键问题展开研究。论文的主要研究和创新性成果如下:首先,在视频质量评价方面,提出一种基于冗余观测值的部分参考CS视频质量评价方法和一种体现主观感知质量特性的客观分层CS视频质量评价方法。前者以客观和部分参考的形式,以较低的附加冗余观测值的成本实现了对CS视频恢复后的质量评价,所获质量信息与PSNR值具有较强的相关性;同时,依据所获视频质量信息,实现CS视频质量信息反馈观测率的自适应调整方案,与固定观测率CS视频传输相比,视频整体质量得到有效提高。后者评价模型分别从观测层、流层和分组层次描述CS视频网络传输参数对视频质量的影响,这种分层模型可根据应用需求提供不同层的CS视频质量信息,而且模型参数通过主观CS视频质量数据回归分析获得;所获结果具有人类主观感受特征,而且由该模型所获的CS视频质量信息与验证对比实验获得的主观CS视频质量信息具有强相关性。其次,在提高编码效率方面,针对CS量化编码,提出一种观测值删除量化方法。该方法从观测值特征分析出发,考虑观测值全局投影和近似高斯分布的性质,通过观测值删除来缩减观测值值域范围,从而在一定的量化电平数量下减少量化误差、提高观测值量化率失真性能。本方法在适合的删除参数下,CS图像重建质量优于直接均匀标量量化方法和分块图像压缩感知DPCM方法。最后,在传输差错控制方面,提出一种基于显著性信息的可分级编码差错控制方法和一种基于单比特奇偶校验的检错删除差错控制方法。前者针对大数据量图像信号单路径传输的低可靠性,依据多径分集和图像显著性分析技术,实现针对CS图像的可分级编码这种不对称信道保障方法;该方法在无差错环境中率失真性能优于传统无显著性信息方法,在丢包环境下率失真性能优于CS多描述编码方法。后者从CS视频传输观测值结构特征出发,以在观测值数据段增加单比特奇偶校验位的检错删除方式实现对CS视频的差错控制,此方法实现简易且参数可根据信道状态调整,可在不同误比特率信道条件下,获得接近BCH码和RCPC码的性能,而且在码率和计算复杂度上具有明显优势。
[Abstract]:In the era of information technology development, the application of image video as the carrier occupies an important position in the information transmission. At the same time, the image video processing and transmission technology has received widespread attention. The mature and efficient prediction video coding method represented by H.264 is the basis of the image video processing, but the prediction class is made up. The code needs a huge amount of computation, and it will cause the channel error to affect the continuity. It is not suitable for some wireless application scenarios such as participatory perception and wireless multimedia sensor network, such as participatory perception and wireless multimedia sensor network (CS). The theory provides a new way of thinking to solve the above problems. The limitation of Nyquist frequency, and the compression of image or video data to a certain extent in the way of dimensionality reduction at the same time of sampling observation, realizes the merging of the two processes of sampling and compression, and avoids the resource consumption produced by the traditional pre sampling compression. Moreover, the compression perception is used as a global observation of information. Information sampling) instead of the traditional signal local sampling (the so-called signal sampling) new signal acquisition mode, the observed values obtained by linear random projection carry the whole information with the same importance, which is beneficial to the construction of a simple and effective anti channel error scheme. Reducing the pressure of high-speed sampling, reducing the cost of node acquisition, coding, processing and transmission, will further promote the development of information field. On the basis of compressed sensing theory, this paper focuses on image / video signal as the main research object, closely surrounding video quality evaluation, coding efficiency and image / video error control. In three aspects, the key issues related to video transmission quality are studied. The main research and innovative results of this paper are as follows: firstly, a partial reference CS video quality evaluation method based on redundant observation values and an objective hierarchical CS video quality evaluation based on subjective perceptual quality characteristics are proposed. In the form of objective and partial reference, the former realizes the quality evaluation of CS video recovery with the cost of lower additional redundant observations. The quality information has a strong correlation with the PSNR value. At the same time, according to the video quality information obtained, the adaptive adjustment scheme of the feedback observation rate of the CS video quality information is realized. Compared with the fixed observation rate CS video transmission, the overall quality of the video is effectively improved. The latter model describes the effect of the CS video network transmission parameters on the video quality from the observation layer, the flow layer and the packet level respectively. This hierarchical model can provide the CS video quality information of different layers according to the application requirements, and the model parameters pass through the subjective CS. The result of video quality data regression analysis is obtained; the results obtained have human subjective feelings, and the CS video quality information obtained by the model has strong correlation with the subjective CS video quality information obtained by the verification contrast experiment. Secondly, in the aspect of improving the coding efficiency, a method of observation value deletion and quantization is proposed for the CS quantization coding. This method, starting from the analysis of observation value features, takes into account the global projection of observation values and the properties of approximate Gauss distribution, and reduces the range of observation value range by observation value deletion, thus reducing the quantization error and improving the distortion performance of the observed value quantization rate under certain quantized level. Under the suitable deletion parameters, the CS image is reconstructed. The quantity is superior to the direct uniform scalar quantization method and the block image compression sensing DPCM method. Finally, in the aspect of transmission error control, a scalable coding error control method based on significant information and a error detection and deletion control method based on single bit parity check are proposed. The former aims at the single path transmission of large data amount image signals. The low reliability of the transmission is based on the multipath diversity and image saliency analysis technology to achieve an asymmetric channel guarantee method for CS images. The rate distortion performance in the error free environment is superior to the traditional non significant information method. In the packet loss environment, the rate loss is superior to the CS multi description coding method. The latter is viewed from CS. The error control of CS video is realized by adding the error detection and deletion method of the single bit parity bit in the observation data segment. This method is simple and parameters can be adjusted according to the channel state. Under the condition of different bit error rate channels, the performance is obtained near the BCH code and RCPC code, and the bit rate and the rate are also available. The computational complexity has obvious advantages.
【学位授予单位】：南京邮电大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TN919.8

【参考文献】