【作者】 李茜；

【导师】 朱宏擎；

【作者基本信息】 华东理工大学 ， 信号与信息处理， 2013， 硕士

【摘要】 为消除图像压缩编码中经常产生的块效应,重叠变换作为一类改进的变换方法在20世纪80年代被提出来,之后关于它的一系列高性能的数据压缩算法,受到了国内外学者们广泛关注与研究。重叠正交变换是重叠变换的基础,调制重叠变换以及复数域重叠变换均是它的改进与提高,也是重叠变换领域中的重要变换。经验表明,图像可稀疏表示,压缩感知图像重建算法正式利用这一先验知识,采用比奈奎斯特采样率低得多的随机投影观测值达到了重建原始图像的目的。这使得它在信号、图像处理领域有着极大的研究价值以及广阔的发展前景。为了消除在高压缩比时传统分块不重叠变换未利用相邻块信息的相关性所导致的块效应,本文着重实现了三种重叠变换：重叠正交变换、调制重叠变换及复数域调制重叠变换。实验结果表明这三种重叠变换的压缩性能均优于传统方法,而且能够在高压缩比时消除明显的块效应。为了克服传统分块压缩感知重建算法未利用变换系数的邻域统计特性的缺点,以及消除通过该方法重建的图像的块效应,本论文还提出了将重叠变换方法应用于传统压缩感知算法中,并就正交匹配追踪法(Orthogonal Matching Pursuis,OMP)和平滑投影Landweber法这两种图像重建算法进行研究。实验结果表明,在同等条件下,平滑投影Landweber法重建效果要优于正交匹配追踪法；基于重叠变换的压缩感知图像重建算法要优于传统分块压缩感知图像重建方法。更多还原

【Abstract】 Lapped transforms is proposed in order to eliminate block effects that appear in the process of image compression coding, it has been concerned and studied extensively for its high performance. Lapped transforms is based on lapped orthogonal transform, lapped modulated transform and modulated complex lapped transform obtained from lapped orthogonal transform, they are also show its significance. Compressed sensing image reconstruction algorithm takes advantage of the feasibility of image sparse representation and reconstructs the original images with the observed value of random projections. It shows great research value and broad prospects for development in signal and image processing field.In this paper, we show three lapped transforms to eliminate block effects generated due to ignoring the usage of correlation of adjacent blocks in the high compression ratio:lapped orthogonal transform, modulated lapped transform and modulated lapped complex transform. The performances of the proposed methods are compared with traditional methods. The experimental study shows that three lapped transforms with better compression performance can eliminate block effects. What’s more, this study proposes a method that use lapped transforms in compressed sensing algorithm to overcome block effects in traditional block-based compressed sensing algorithm and disadvantage of ignoring the usage of statistical characteristics of the neighborhood transform coefficients. This paper also discuss orthogonal matching pursuit method and smooth projection Landweber method.The experimental results show that smooth projection Landweber method performs better than orthogonal matching tracing method under the same conditions, and compressed sensing image reconstruction algorithm based on lapped transforms outperforms traditional methods.更多还原