一种改进的基于稀疏表示的超分辨率重建算法

An Improved Algorithm of Image Super-Resolution Reconstruction Based on Sparse Representation

针对超分辨率重建中出现的图像模糊等问题,提出了一种改进的基于稀疏表示的超分辨率重建算法。算法将局部线性嵌入(LLE)引入到稀疏编码中,从而保证稀疏分解的稳定性,能够有效地保留数据的局部流形结构。而在字典训练阶段和重建阶段,采用基于图像块清晰度的聚类方法。对于每一类图像,训练一对低和高分辨率的字典,使得字典继承该类图像的清晰度,提高了学习性能。实验结果表明,改进的算法能够有效地提高重建的高分辨率图像的精度和效果,提高了峰值信噪比。

In order to solve the problem of image blur in super resolution reconstruction, we propose an improved algo- rithm of image super- resolution reconstruction which is based on representation. By incorporating the LLE based mani- fold learning as a regularizer into the traditional sparse coding objective function, the proposed method can effectively keep the local manifold structure of data. Patch clustering in the dictionary training stage and model selection in the reconstruc- tion stage are based on patch sharpness. For each cluster, a pair of coupled low- and high- resolution dictionaries is learned. The designed cluster dictionaries are shown to inherit the sharpness of their respective clusters, and the learning performance is improved. Experiments show that compared with current a couple kinds of algorithms, the proposed meth- od can improve the accuracy and effectiveness of the reconstruction of high resolution images and improve the PSNR value.