Online dictionary learning super-resolution reconstruction method based on sparse representation

Abstract

The invention relates to an online dictionary learning super-resolution reconstruction method based on sparse representation. The online dictionary learning super-resolution reconstruction method comprises the three parts of prior information, dictionary learning and sparse reconstruction. The dictionary training part adopts online dictionary learning, so that not only is external image library information effectively utilized, but also the image information is added to update the dictionary. Besides, on the basis of sparse prior, a local autoregression model and non-local self-similarity areadded at the same time to serve as prior information, a non-local regularized super-resolution reconstruction model is established, and various structural features of the image are reconstructed. In the sparse reconstruction stage, a sparse coefficient is determined by using multi-scale self-similarity sparse representation, a non-local constraint item is constructed according to a corresponding relation between different scale similar blocks, and additional information of a multi-scale self-similarity structure is introduced into a reconstruction process in an image reconstruction model; Themethod not only can reduce the dependence of the test image on the training image set, but also can overcome the local distortion or fuzziness of the image block in the reconstruction process, therebyfurther improving the quality of the reconstructed image.

Description

technical field [0001] The invention relates to the technical field of image signal processing, in particular to a method for super-resolution reconstruction based on online dictionary learning with sparse representation. Background technique [0002] There are many ways for human beings to understand the world and obtain information, including auditory, tactile and olfactory, and visual information is also indispensable. The importance of visual information is often the most important, far surpassing other ways of acquisition. Therefore, the image quality obtained through visual information will directly affect our cognition and judgment of things. [0003] High-resolution images have higher pixel density and contain more detailed information, which is of great significance and help to various special fields such as image feature extraction, target recognition and positioning, satellite remote sensing imaging, and medical imaging. However, in practice, the obtained image ...

AI Technical Summary

Problems solved by technology

At present, the traditional super-resolution reconstruction method has poor reconstruction effect, limited ability to maintain details such as edges and textures, and is prone to visual artifacts.

Images