Image Super-Resolution Reconstruction Based on Online dictionary learning Algorithm

Abstract

For the image super-resolution reconstruction method based on case-learning shows that a fast and efficient dictionary learning algorithm is very important to solve the problem of mapping inconsistency between low-resolution and high-resolution images. This paper adopts the online dictionary learning algorithm for the image super-resolution. In the learning stage, the algorithm constructs the high-resolution and the corresponding low-resolution feature training sets, then by using the online dictionary learning algorithm, obtains a sparse coding matrix of the low-resolution training sets, and computers the high-resolution dictionary by sharing the sparse coding coefficients; in the reconstruction stage, the input low-resolution image firstly is interpolated to the size of the desired high-resolution image, and obtains the sparse coding matrix through OMP ( Orthogonal Matching Pursuit ) method in the low-resolution test sets, then computers the high-resolution image blocks based on the above high-resolution dictionary and the later sparse coding matrix, finally reorders and averages the blocks to achieve the reconstructed high-resolution image. The experimental results show that the proposed method can achieve better quality for image super-resolution reconstruction than the traditional sparse coding method, the detail and texture of the reconstructed image are reconstructed well, and the algorithm can effectively inhibit the artifact of image edge phenomenon.