Super-resolution processing method for image based on blind fuzzy estimation and anchoring space mapping

Abstract

The precision and the timeliness of a training sample are two challenges in current image super-resolution field. The invention provides a super-resolution processing method for a single image with respect to the problem. The method comprises a fuzzy kernel function estimation stage and a super-resolution reconstruction stage. The fuzzy kernel function estimation is realized by minimizing image cross-scale diversity, to improve the precision of the training sample; and the super-resolution reconstruction is realized by anchoring space mapping on the basis of a feature extraction strategy of reverse projection residuals and dual dictionary learning, to enhance the expression capability of an algorithm on the bottom feature of the image and accurately reflect the mapping relationship between low-resolution and high-resolution feature spaces. To improve the efficiency of the algorithm, the high-resolution feature reconstruction is completed by adopting anchoring space mapping, and both the fuzzy kernel function estimation stage and the super-resolution reconstruction stage adopt a block processing manner. The method provides a powerful competitive advantage for improving the performance of imaging equipment, and provides necessary theoretical and technical supports for further improving the super-resolution processing effect.

Description

Technical field: [0001] The present invention relates to image / video processing technology, specifically, the present invention relates to a single image super-resolution processing algorithm for blind blur estimation and anchor space mapping. technical background: [0002] Image super-resolution is the technique of generating a high-resolution image of the same scene from one or more low-resolution images of the same scene. In essence, the super-resolution problem is a serious ill-conditioned problem, which can only be solved with some prior information. From the perspective of the historical development of this field, super-resolution techniques can be roughly divided into three categories: interpolation-based methods, reconstruction-based methods, and machine learning-based methods. Interpolation-based methods have been developed very maturely. The mainstream interpolation algorithms include nearest neighbor interpolation, bilinear interpolation, bicubic interpolation, c...

AI Technical Summary

Problems solved by technology

[0005] The purpose of the present invention is to solve the problems that the training data used in the existing image super-resolution algorithm based on machine learning is not high in accuracy, incompatible with test samples, and too time-consuming to meet practical applications.

Images