视频压缩感知和ARMA融合研究

发布时间：2018-05-29 04:19

本文选题：视频模型 + 压缩感知　；参考：《合肥工业大学》2014年硕士论文

【摘要】：压缩感知理论及其应用是当前信息领域科学研究的前沿性课题之一,涉及到数学、物理及计算机科学等诸多学科和优化、信号估计等众多复杂的问题。由于压缩感知技术可以大大降低数据采集量,加上视频处理过程一直被维数问题所困扰,故压缩感知在视频处理方面的应用占有重要地位。针对视频信号本身结构的复杂性,基于常规的压缩感知建模方法基本上是从图像的压缩感知过渡过来的,一直沿用着对视频帧一帧一帧单独处理的思路,其在实际的视频处理中优势并不明显,该文将传统的压缩感知视频处理模型和自回归滑动平均模型(ARMA)相结合,提出了将这两个模型相融合的思路。本文从传统视频压缩模型和压缩感知视频模型所面临的问题出发,做了三方面工作和创新：1、在研究了传统视频压缩感知模型和ARMA视频模型的优缺点的基础上,结合了二者的优点,建立一种新的融合了压缩感知和ARMA模型的视频模型；2、在视频重建过程中,充分利用视频帧内稀疏性和帧间相关性,把视频分割成动态部分和静态部分同时采样但分别处理,同时给出了这种新模型在理论及实践上的鲁棒性证明；3、该模型针对传统的压缩感知重建算法噪声问题,引入了noiselet变换降噪,改进了Model Based Cosamp算法求解混合范数问题,最终得到视频状态空间ARMA模型的关键参数。在多组真实场景下的实验结果表明,该模型较大程度上降低了帧间冗余度和数据采集量。在该模型下,视频采集压缩比在100至200时,仍然能取得较好的重建效果。
[Abstract]:The theory of compressed perception and its application is one of the leading subjects in the field of information science, involving many complicated problems, such as mathematics, physics and computer science, optimization, signal estimation and so on. Because compressed sensing technology can greatly reduce the amount of data acquisition, and the video processing process has been troubled by the dimension problem, compression sensing plays an important role in the application of video processing. In view of the complexity of the video signal itself, the conventional compressing perceptual modeling method is basically transferred from the compression perception of the image, and has always followed the idea of processing a frame of a video frame separately. This paper combines the traditional compressed perceptual video processing model with the autoregressive moving average model (ARMA) and puts forward the idea of merging the two models. Starting from the problems of the traditional video compression model and the compressed perceptual video model, this paper has done three aspects of work and innovation: 1. On the basis of studying the advantages and disadvantages of the traditional video compression perception model and the ARMA video model, Combining the advantages of the two models, a new video model, which combines compression perception and ARMA model, is established. In the process of video reconstruction, the inner-frame sparsity and inter-frame correlation are fully utilized. The video is divided into dynamic part and static part and sampled simultaneously but processed separately. The robustness of the new model is proved theoretically and practically. The model is aimed at the noise problem of the traditional compression perceptual reconstruction algorithm. The noiselet transform is introduced to reduce the noise, and the Model Based Cosamp algorithm is improved to solve the mixed norm problem. Finally, the key parameters of the video state space ARMA model are obtained. The experimental results in multiple real scenes show that the model can reduce inter-frame redundancy and data acquisition to a large extent. In this model, the video capture compression ratio of 100 to 200 can still achieve good reconstruction results.
【学位授予单位】：合肥工业大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN911.73

【相似文献】