基于高频和中频信息的图像超分辨率重建

Image Super-resolution Reconstruction Based on High-and Mid-frequency Components

提出一种基于图像高频和中频信息的超分辨率算法。分别将图像高频和中频作为高分辨图像和低分辨率图像的特征,以图像高频和中频信息作为训练样本对,采用全局迭代收缩方法(GISA)进行稀疏分解,获得高、中分辨率字典对。根据测试图像对应的中频信息和字典对获得图像高频信息,结合测试图像插值放大结果,经非局部相似性方法处理后获得高分辨率图像。实验结果表明,提出的方法具有较高的重建质量。

An image super-resolution（ SR） reconstruction algorithm based on high- and mid-frequency components is proposed in this paper. The algorithm selects high- and mid-frequency of natural images as the feature of high-resolution（ HR） images and low-resolution（ LR） images respectively. Patch pairs,composed of high- and mid-frequency components,are trained by GISA（ generalized iterated shrinkage algorithm） to obtain the high-resolution and mid-resolution joint dictionary pair. According to the mid-frequency of the test image and the dictionary pair,image high-frequency is reconstructed. Then,combined with interpolated LR images and reconstructed high-frequency,the HR image is reconstructed after a non-local similarity regularization term. Experimental results show that the proposed method has better performance.