Super-resolution method based on single image

Abstract

The invention relates to a super-resolution method based on a single image. The super-resolution method includes the steps that S1, bicubic interpolation is carried out on the input low-resolution image to obtain an initial high-resolution image; S2, the initial high-resolution image is divided into a plurality of image blocks overlapped mutually, then a similar image block grouping is obtained, and image noise of the similar image block grouping is removed; S3, the multiple denoised image blocks are fused into a whole high-resolution image, a non-local similar image block and a weighting coefficient of each image block are solved, and the redundancy weight of a non-local similar image block grouping is calculated; S4, an on-line dictionary is updated according to the similar image block grouping and fused with an off-line dictionary; S5, the sparse representation coefficient, about a fused dictionary, of each image block is solved; S6, all the image blocks and the whole high-resolution image are reconstructed, if iterations do not converge and the number of the iterations is smaller than a preset threshold value, the previous steps are executed again, and otherwise the high-resolution image is output. The reality sense and accuracy of super-resolution reconstruction are promoted, and the super-resolution method has the advantage of removing the image noise at the same time.

Description

technical field [0001] The invention relates to the technical field of computer image processing, in particular to a method for super-resolution based on a single image. Background technique [0002] At present, image super-resolution reconstruction is an important research topic in the field of computer image processing, which is widely used in video surveillance, satellite remote sensing imaging, medical images and other fields. Generally, the image degradation model can be expressed as: [0003] y=DHx+v (1) [0004] where y is the degraded low-resolution image, x is the original high-resolution image, D and H are the downsampling matrix and blur matrix, respectively, and v is the additive white Gaussian noise. The super-resolution methods of a single image are mainly divided into three categories: the first category is the interpolation method, which is simple and fast but is prone to image over-smoothing and jagged effects; Fidelity term for the degradation model On...

AI Technical Summary

Problems solved by technology

[0013] The technical problem to be solved by the present invention is how to provide a method that can effectively remove image noise, especially non-Gaussian noise, and improve the adaptability and robustness of image self-similarity regularization and dictionary learning in view of the deficiencies in the prior art key issues

Images