Single-image super-resolution method

Abstract

The invention discloses a single-image super-resolution method. The method comprises the following steps: up-sampling a single low-resolution image to obtain an image P1; inputting the P1 into a VDSR,and outputting a high-resolution image P2; performing Saak transformation on the P1 and the P2 by using the transformation kernels T1 and T2 respectively, and obtaining 2n2 Saak feature maps by usingthe P1 and the P2 respectively, wherein the transformation kernel T1 and the transformation kernel T2 are respectively calculated by P1 and P2, the convolution kernel of Saak transformation is n * n,and n is a natural number; selecting the first m Saak feature maps from the 2n2 Saak feature maps of the P1 as a training set to train a convolutional neural network, inputting the selected m Saak feature maps into the convolutional neural network, and outputting the m feature maps of the P1, 1 < = m < = n2; selecting the last 2n2-m Saak feature maps from the 2n2 Sak feature maps of P2 to form 2n2 maps with the m feature maps of P1 output by the convolutional neural network, and performing Saak inverse transformation to obtain a high-resolution image P3; and fusing P2 and P3 to obtain a finalhigh-resolution image P4.

Description

technical field [0001] The invention relates to the fields of computer vision and digital image processing, in particular to a single image super-resolution method. Background technique [0002] Single image super-resolution reconstruction technology refers to an image processing technology that uses a computer to process a low-resolution image to restore a high-resolution image. It has always been a very popular direction in the field of computer vision. High resolution means that the image has a high pixel density, which can provide more details, which often play a key role in the application. Therefore, it is widely used in video surveillance, target detection, face recognition, automatic driving and other fields. [0003] Due to the asymmetry of high- and low-resolution image information, this is a typical ill-posed problem, that is, a low-resolution image often corresponds to countless high-resolution solutions. In the current super-resolution algorithm, whether it is...

AI Technical Summary

Problems solved by technology

[0006] Image quality evaluation index is an important part of evaluating super-resolution image quality. It is mainly divided into two categories: objective evaluation and subjective evaluation. The commonly used objective evaluation indexes include mean square error (MSE), peak signal-to-noise ratio (PSNR) and structural similarity Sex Index (SSIM), but objective evaluation indicators often cannot truly reflect human subjective visual effects

Subjective evaluation mainly relies on methods such as manual scoring and averaging, which can effectively reflect the quality of visual effects, but due to different subjective feelings, the evaluation results often vary from person to person

Images