基于多通道盲复原和改进K-SVD模型的图像恢复

Image Restoration Based on Multi-channel Blind Restoration and Sparse Representation Method

图像盲复原(IBR)问题一直是图像处理中的重要研究课题。目前空间不变的多通道图像盲复原算法研究较为普遍,这种算法具有较好的盲去模糊效果,但是对噪声的抑制能力不足,特别是对含有大量噪声的低分辨率图像而言,消噪效果较差。基于K-奇异值分解(K-SVD)的模型能够有效地处理噪声方差较大的图像,但是不能自适应图像的稀疏先验性。为了解决上述问题,在全变分(TV)多通道IBR算法处理的基础上,结合一种改进的K-SVD消噪模型的优势,提出了一种新的组合图像恢复方法。改进的K-SVD模型考虑了图像特征系数的稀疏先验知识和最大化稀疏度,具有自适应的消噪鲁棒性。分别采用模拟的和真实的低分辨率图像(毫米波图像)进行测试,与采用单一的多通道盲恢复和图像消噪算法相比,实验结果表明所提出的图像恢复方法具有较好的视觉效果和较高的信噪比。

The problem of image blind restoration （ IBR） has been the important research issue in image process-ing. At present, the research of multi-channel with space invariant is very common. This algorithm behaves certain advantages in blind de-blurring, but it is limited to denoise images. Especially, to low resolution （ LR ） images, which contain much unknown noise, the restored effect is worse only using the multi-channel restoration technique. K-mean based singular value decomposition （ K-SVD） model can denoise images with large noise variance, however, it is not self-adaptive to an image’ s priors. To solve this defect, on the basis of processed results by the total variation （ TV） based multi-channel restoration algorithm, further combined a modified K-SVD model’ s advantage in denoising images, a novel combined image restoration method is discussed here. This modified K-SVD model considers the sparse priors and maximize sparsity of image feature coefficients and behaves self-adaptively denoising robust. In test, the artificial LR image and real LR image （ i. e. millimeter wave image） are used, and compared with the single multi-channel IBR method and the single image denoising method, experimental results show that our method behaves bet-ter visual effect and higher values of signal noise ratio.