基于低秩约束和边信息的近似消息传递CS重构算法

噪声环境下图像压缩感知(compressive sensing,CS)重构方法的性能会大幅度下降。在近似消息传递(approximate message passing,AMP)算法的基础上,同时利用结构先验信息和边信息来增强AMP算法对噪声的鲁棒性。利用图像中相似块的低秩特性,在反投影的含噪图像中捕获低秩子空间的结构特征;再将含有确定成分的前期重构图像作为边信息，以实现细节的增强。实验表明,本文算法比原始AMP算法在峰值信噪比(peak signal to noise ratio,PSNR)上平均提高了3.89 dB,且获得更加清晰的重构图像;与仅利用低秩特性的AMP算法相比,引入边信息后本文算法在PSNR上获得了0.27 dB的增益,同时增强了重构图像的细节。

Approximate message passing based on low-rank constraint and side information for CS recovery

The performance of image compressive sensing (CS) recovery algorithms degrades seriously in the presence of noise. Structured prior information and side information are simultaneously utilized to enhance the robustness of the approximate message passing (AMP) algorithm to noise. The low-rank property of similar patches is exploited to capture low-rank subspace structures. The reconstructed image in the previous iteration that contains some identified components is taken as side information to enhance details. Comparing with the original AMP algorithm and the proposed method without the aid of side information, the proposed method averagely improves peak signal to noise ratio (PSNR) by 3.89 dB and 0.27 dB while producing clearer and more detailed images respectively.