基于分布式压缩感知的视频编解码研究

发布时间：2019-01-17 18:11

【摘要】：近年来,计算机技术发展迅速,处理能力日新月异,这也给互联网、视频及电视广播领域带来了改革之风。作为信息的重要载体,视频技术也随之不断发展进步。视频编码一直是视频处理领域的热点研究课题,视频的高分辨率、立体化发展导致视频数据日益增加,高效的压缩编码技术显得至关重要。压缩感知理论作为一个全新的采样理论,撼动了信息处理领域的基石——奈奎斯特采样定理。压缩感知理论可以在不丢失重构原信号所需信息的情况下,采用远低于传统方法的采样频率,用最少的观测次数来采样信号,实现信号的降维处理。分布式压缩感知理论结合了压缩感知理论与分布式信源编码理论,可以充分利用信号内及信号间的相关性,受到无线传感网络(WSN)、雷达及视频编码等领域广泛关注。本文主要针对分布式压缩感知视频编解码进行研究,所做的工作如下:首先,本文对压缩感知重构算法及分布式压缩感知理论进行了研究,然后针对第一联合稀疏模型JSM1提出了一种联合重构算法,该算法充分利用信号之间的相关性,在减少测量值数目的情况下,加速了信号的重构,并保证了信号重构的精确度。然后,本文基于视频特性,结合联合稀疏理论和分布式视频技术,对基于分布式压缩感知的视频编码技术进行了研究。针对视频编码数据量大,采集耗时等特点,引入分布式压缩感知,提出了一种结构简单,处理高效的视频编解码框架。在本框架中,采用联合稀疏模型减少计算量,在提供快速编解码的同时,实现了图像的精确重构。最后,本文对图像稀疏模型进行了研究,结合视频联合稀疏的特点,将树稀疏模型引入分布式压缩感知中,提出了森林稀疏模型,该模型充分挖掘视频图像帧的树稀疏结构及联合树稀疏结构特点,能够较快的解码视频图像。实验结果表明,该模型下的视频重构更为简单稳定。综上所述,本文提出的两种分布式压缩感知视频编码方案的性能要明显优于独立使用压缩感知进行编解码的视频编码方案的性能。
[Abstract]:In recent years, the rapid development of computer technology, processing power with each passing day, which has brought the Internet, video and television broadcasting reform. As an important carrier of information, video technology has been developing and improving. Video coding has always been a hot research topic in the field of video processing. The high resolution and three-dimensional development of video leads to the increasing number of video data. Efficient compression coding technology is very important. As a new sampling theory, compressed sensing theory has shaken the cornerstone of information processing field-Nyquist sampling theorem. The compressed sensing theory can use the sampling frequency far lower than the traditional method without losing the information needed to reconstruct the original signal and sample the signal with the least number of observations to realize the signal dimensionality reduction processing. The theory of distributed compression sensing combines the theory of compression sensing with the theory of distributed source coding, which can make full use of the correlation between signals and signals. It has attracted wide attention in the field of (WSN), radar and video coding in wireless sensor networks. This paper mainly focuses on the research of distributed compressed perceptual video coding and decoding. The work is as follows: firstly, this paper studies the algorithm of compressed perception reconstruction and the theory of distributed compressed sensing. Then a joint reconstruction algorithm is proposed for the first joint sparse model (JSM1). The algorithm makes full use of the correlation between signals, accelerates the reconstruction of signals and ensures the accuracy of signal reconstruction under the condition of reducing the number of measured values. Then, based on the characteristics of video, combined with sparse theory and distributed video technology, this paper studies the video coding technology based on distributed compression perception. In view of the characteristics of large amount of video coding data and time-consuming collection, a simple and efficient video coding and decoding framework is proposed by introducing distributed compression perception. In this framework, the joint sparse model is used to reduce the computation cost, and the accurate reconstruction of the image is realized at the same time of fast coding and decoding. Finally, the sparse image model is studied in this paper. Combined with the characteristics of video sparsity, the tree sparse model is introduced into the distributed compression awareness, and the forest sparse model is proposed. The model fully exploits the tree sparse structure of video frame and the sparse structure of joint tree, which can decode the video image quickly. The experimental results show that the video reconstruction based on this model is simpler and more stable. To sum up, the performance of the two distributed compressed perceptual video coding schemes proposed in this paper is obviously superior to that of the video coding schemes using compression awareness independently.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN919.81

【参考文献】