基于软阈值贪婪方向追踪的图像重构法

Image Reconstruction Method Based on Soft-Thresholding Greed Direction Pursuit

首先,使用新型压缩感知框架对图像进行基于小波的轮廓波变换(WBCT);然后,采用部分傅里叶矩阵随机观测子空间系数;最后,结合软阈值贪婪方向追踪(SGDP)算法完成图像重构。试验结果表明,与传统的稀疏梯度投影(GPSR)算法相比,基于软阈值贪婪方向追踪的图像重构法在恢复图像低频信息的同时保留了高频信息,尤其对于含噪图像重构,可使重构图像峰值信噪比(PSNR)值提高约2dB。

Firstly, a wavelet-based contourlet transform （WBCT） is done with the image by the new compressed sensing framework. Then, the subspace coefficient is observed by the partial Fourier matrix. Finally, the image is reconstructed through the soft-thresholding greed direction pursuit （SGDP） algorithm. The experimental result shows that the method can reconstruct the low frequency information of image as well as hold the high frequency information compared with the traditional gradient projection for sparse reconstruction （GPSR） algorithm. Particularly for the reconstructed noisy image, the method can enhance about 2 dB of the peak signal to noise ra-tio （PSNR） on the reconstruction image.