基于稀疏表示的单帧运动图像盲复原

摘要

This paper presented a blind motion deblurring method based on sparse representation which made full use of nature image priors. It contained two phases; kernel estimation and image restoration. In the step of kernel estimation, it used shock filter to predict sharp edges from blurring image which conducted global image restoration. It applied multi-scale scheme to solve big kernel problem. In the step of image restoration, it applied sparse representation to denoise and rebuild latent image which improved image restoration quality. Experimental results show that the proposed method can effectively remove motion blurring under different noises and kernels.%提出一种基于稀疏表示的单帧运动盲复原方法,它充分利用自然图像中存在的各种先验知识进行求解.该方法分为模糊核估计和图像修复两个阶段.在估计模糊核时,它运用shock滤波器从模糊图像中预测出清晰边缘,以此指导全局图像的复原,并运用多尺度策略来解决大模糊核问题.在图像修复阶段,运用稀疏表示理论对复原图像进行降噪和重建,最终提高图像复原质量.实验结果表明,在不同噪声和模糊核条件下,该算法能有效消除运动模糊.