一种分块自适应压缩感知图像重构算法

Image Reconstruction Algorithm for Compressed Sensing Based on Discrete Cosine Transform

CS理论中,在离散余弦变换下使用OMP算法重构图像时需要较高的测量值可以获得较好的重构效果,但是存在重构图像模糊的问题。为此,提出了基于离散余弦变换的图像分块自适应正交匹配追踪(BAD-OMP)算法。基于分块压缩感知技术,对图像进行均匀分块处理,根据图像块稀疏性进行自适应采样,再用均值滤波算法平滑处理,从而减少重构所需的测量值,降低块效应。仿真结果表明,采样率取0.1~0.35时,BAD-OMP算法重构图像的PSNR值较OMP算法的PSNR值高9~11 d B,实现了在低采样率下获得较高的重构质量。

In the CS theory,the use of OMP algorithm to reconstruct the image under discrete cosine transform requires a higher measurement value to obtain a better reconstruction effect,but there is a problem of reconstructing the image blur. In this paper,a Block-based Adaptive based on Discrete cosine transform with OMP（ BAD-OMP） algorithm is proposed. Based on the block compression sensing technique,the image is uniformly punctured,and the adaptive sampling is carried out according to the sparseness of the image block,and the mean filtering algorithm is used to smooth the processing,thus reducing the required measurement value and reducing the blocking artifacts. The simulation results show that the PSNR value of the reconstructed image of BAD-OMP algorithm is 9 ~ 11 d B higher than the PSNR value of OMP algorithm,and the higher reconstruction quality is achieved at low sampling rate.