Image super-resolution method based on dictionary learning and non-local total variation

Abstract

The invention discloses an image super-resolution method based on dictionary learning and non-local total variation. The image super-resolution method based on the dictionary learning and the non-local total variation mainly solves the problems of ring effect, lost high-frequency information, and inaccuracy of boundary matching of a super-resolution method and the like. The achieving steps include (1), inputting an image training set; (2) using a KSVD algorithm to train two corresponding high-resolution dictionary and a low-resolution dictionary; (3) conducting sparse representation of a low-resolution input image, and determining sparse coefficient; (4) using determined sparse coefficient and the high-resolution input image to obtain a high-resolution image; (5) conducting ring effect removal of the non-local total variation on the reconstructed high-resolution image; (6) conducting high-frequency information enhancing of the high-resolution image by error compensation, and obtaining a final result. By showing of a simulation experiment, compared with the prior art, the image super-resolution method based on the dictionary learning and the non-local total variation has the advantages of being simple in operation, small in noise, clear in edge and the like, and can be used for obtaining a high-definition image.

Description

technical field [0001] The invention belongs to the technical field of image processing, and relates to an image super-resolution method, in particular to an image super-resolution method based on dictionary learning and non-local total variation, which can be used to improve the resolution of natural images. Background technique [0002] Image resolution is an important indicator to measure image quality. With the invention of charge-coupled devices and complementary metal-oxide-semiconductor image sensors, people have made some progress in the quality of acquired images, but image sensors are prone to blurring, undersampling, noise, etc. when acquiring images. Factors, so the image quality is difficult to further improve. People hope to obtain high-resolution images by improving the image sensor from the hardware aspect, but the cost of this method is too expensive to be popularized. Therefore, it was proposed to use image super-resolution method to improve the resolutio...

AI Technical Summary

Problems solved by technology

For example, Freeman et al. proposed a learning-based image super-resolution method, which uses Markov random fields and prior knowledge to learn the relationship between low-resolution images and high-resolution images, and then reconstructs high-resolution images. Resolving images, but this method cannot well maintain the boundary information of high-resolution images, the computational complexity is large, and the efficiency is low

[0005] In the document Image super-resolution as sparse representation of raw image patches, Yang et al. of the University of Illinois in the United States proposed to use dictionary learning and sparse representation theory to realize the image of a single frame image. Super-resolution, in this method, due to the sparse representation over-reliance on the constructed over-complete dictionary and the defects of its dictionary learning algorithm, the edge of the obtained high-resolution image has a ringing effect, and the edge and texture of the high-resolution image are not clear enough or even the real edge Contradictory textures, loss of high-frequency information, etc.

Images