基于复合稀疏约束的近似消息传递CS重构算法

发布时间：2018-10-17 08:42

【摘要】：压缩感知(CS)重构中的近似消息传递(AMP)算法通过迭代执行小波阈值操作和残差更新来快速准确地实现稀疏信号重构,但它所采用的小波系数稀疏约束并不适用于非稀疏的自然图像,尤其CS观测过程存在噪声干扰时.为此,文中提出了一种基于复合稀疏约束和AMP框架的CS图像重构算法,使用相似图像块低秩约束和双边滤波约束作为自然图像的联合先验信息,以改善图像规则纹理和边缘的恢复效果,从而提升算法的重构性能.无噪CS观测的重构实验表明,文中算法的峰值信噪比(PSNR)比仅用低秩约束的AMP算法提高了0.45 d B,比原始AMP算法高6.19 d B;而在含噪CS观测的重构实验中,对应的PSNR增益则分别是0.25和4.60 d B;无论是无噪观测还是含噪观测,文中算法都获得了更佳的主观视觉效果.
[Abstract]:The approximate message passing (AMP) algorithm in compressed perceptual (CS) reconstruction implements sparse signal reconstruction quickly and accurately by iterating through wavelet threshold operation and residual update. However, the sparse constraint of wavelet coefficients is not suitable for non-sparse natural images, especially in the presence of noise interference in the CS observation process. In this paper, a new CS image reconstruction algorithm based on complex sparse constraints and AMP framework is proposed, which uses similar image blocks with low rank constraints and bilateral filtering constraints as joint prior information of natural images. In order to improve the effect of image texture and edge restoration, the reconstruction performance of the algorithm is improved. The reconstruction experiments of noiseless CS observations show that the peak SNR (PSNR) of the proposed algorithm is 0.45 dB higher than that of the AMP algorithm with only low rank constraints, and 6.19 dB higher than that of the original AMP algorithm, while in the reconstruction experiment with noisy CS observations, the PSNR increases by 0.45 dB, and 6.19 dB higher than that of the original AMP algorithm. The corresponding PSNR gain is 0.25 dB and 4.60 dB, respectively.
【作者单位】：华南理工大学电子与信息学院;
【基金】：国家自然科学基金资助项目(61471173)~~
【分类号】：TP391.41

【相似文献】