基于联合稀疏近似的彩色图像超分辨率重建

Super-resolution reconstruction for color images based on simultaneous sparse approximation

针对常用的超分辨率(SR)算法中彩色图像的处理会出现彩色信息的丢失或处理结果色彩偏差较大的问题,提出基于联合稀疏近似(SSA)的彩色图像SR重建方法(SR-SSA)。将多通道数据进行联合稀疏编码(SC),并保证它们具有相同的稀疏性模式;同时考虑了彩色图像的各通道数据,并兼顾了它们之间的相关性,增强了先验知识的表达能力。本文方法有效地将高、低分辨率彩色图像特征块统一进行SC,建立它们之间的稀疏关联,并将这种关联作为先验知识指导图像的SR重建。通过自然图像实验,与其它常用的SR算法对比,SR效果有较好改善。

In this paper, a novel super-resolution （SR） method for color images based on simultaneous sparse approximation （SSA） is presented to overcome the problems of color information loss and perceptible color deviation in common super-resolution （SR） reconstruction methods. The multichannel data is unified for simultaneous sparse coding to make sure the same sparsity patterns. The expression ability of prior information is enhanced because of considering all channels and the correlation among them. Our method efficiently establishes sparse association between high resolution （HR） and low resolution （LR） image feature patches, and uses the association as the priori knowledge to guide SR reconstruction based upon a learned dictionary. Experiments with natural images show that our method is better than several other learning-based super-resolution algorithms.