Single-frame image super-resolution reconstruction method based on manifold regularized sparse support regression

Abstract

The invention discloses a single-frame image super-resolution reconstruction method based on manifold regularized sparse support regression. The method comprises the steps that a high-resolution image block set and a low-resolution image block set are established to serve as a high-resolution image block dictionary and a low-resolution image block dictionary respectively; an input low-resolution image is divided into a plurality of image blocks; the image blocks are subjected to sparse encoding by the low-resolution image block dictionary; a support set is obtained; a neighbor relation of a high-resolution image block support set is computed and remained to a reconstructed high-resolution image block space; a mapping relation between a low-resolution image block space and the high-resolution image block space is learnt; and all high-resolution image blocks corresponding to input low-resolution image blocks are obtained by using the mapping relation, and fused into a high-resolution image. The method adopts a manifold regularized sparse support regression representation model, adaptively selects a sparse representation support set, and utilizes a manifold structure of high-resolution image blocks in the support set to restrain reconstruction of the high-resolution image blocks, so that the higher-quality high-resolution image is obtained.

Description

technical field [0001] The invention relates to the field of image super-resolution, in particular to a single-frame image super-resolution reconstruction method based on popular regular sparse support regression. Background technique [0002] With the development of computer networks and handheld mobile devices for photography, images and videos are increasingly used in our lives. However, due to the limitations of network bandwidth and server storage and other issues, most of the images we obtain are of low resolution and quality, which is far from meeting people's needs. Image super-resolution is a technology that can use image processing algorithms to enhance the details of low-resolution images. It can provide us with high-resolution images that contain more details without requiring more demanding hardware devices. [0003] According to the number of input low-resolution images, super-resolution techniques can be divided into two categories: super-resolution technique...

AI Technical Summary

Problems solved by technology

However, the underlying assumption that high and low resolution image patches have "same sparse representation" is difficult to achieve in practice

Tang et al. in literature 4 (Y.Tang, P.Yan, Y.Yuan, and X.Li, "Single-image super-resolution via local learning," Int.J.Mach.Learn.&Cyber., vol.2 , pp.15–23, 2011.) proposed a local learning regression (LLR) method, which also selects the nearest K sample points from the high and low resolution training set to learn a mapping, but does not take into account the image block The geometric structure information of the manifold space, however, the manifold structure is crucial for image representation and analysis

Images