改进的基于稀疏表示的图像超分辨率重建算法

摘要

In order to improve the resolution of images with single frame degradation,we propose an improved algorithm of image super-resolution sparse reconstruction which is based on over-complete dictionary.The algorithm mainly introduces the idea of joint training to the process of dictionary training to ensure that the image patches with high or low resolution have same representation coefficients as their corresponding over-complete dictionaries,and in image reconstruction process,an iterative back-projection algorithm is enforced to strengthen the global reconstruction constraints.Experiments show that compared with current a couple kinds of algorithms,the improvement of the images reconstructed by this algorithm is distinct in both PSNR and structural similarity.In addition,the new algorithm can be applied to super-resolution reconstruction of single-frame blurred image,and effectively enhance the resolution of the image.%为改善单帧退化图像的分辨率，提出一种改进的基于超完备字典的图像超分辨率稀疏重构算法。该算法主要在字典训练过程中引入联合训练的思想以确保高、低分辨率图像块在其对应的过完备字典上具有相同的表示系数，并在图像重建过程中，利用迭代反投影加强全局重建约束。实验表明，与现有的几类算法相比较，该算法的重建图像无论在峰值信噪比还是结构相似性上均有明显提高。并且可应用于单帧模糊图像的超分辨率重建，有效地提高了图像的分辨率。