How to reconstruct the original image from fewer observations is still a crucial question in compressed imaging. According to the probability distribution characteristics of the random projection energy, a novel adaptive sampling method and the corresponding reconstruction algorithm are proposed. The algorithm makes full use of the priors of the sparse representation based on the dictionary and the non-local properties. In order to achieve the sparse image representation, we construct the redundant dictionary that contains several directional dictionaries and one orthogonal DCT dictionary, and solve the sparse optimization problem with constraint of l1 norm. The proposed compressed imaging algorithm which combines the local traits of the image patches and the non-local properties of the image can reconstruct the high quality image in low sampling rate.%如何以较少的观测值重构出高质量的图像是压缩成像系统的一个关键问题.本文根据图像块随机投影能量大小分布特点,提出了一种新的自适应采样方式以及针对自适应采样的有效重构算法.重构时利用了图像在字典下的稀疏表示原理和图像的非局部相似性先验知识.为实现图像的稀疏表示,文中构造了由多个方向字典和一个正交DCT字典组成的冗余字典,并用l1范数作为约束条件求解稀疏优化问题.由于充分利用了图像块的局部特性和图像的非局部特性,本文的压缩成像算法在低采样率下能重构出较高质量的图像.
展开▼
