Super-resolution image reconstruction method, device and equipment

Abstract

The super-resolution image reconstruction method comprises the steps of generating images with different perception qualities through different super-resolution image generation methods; Calculating perception quality evaluation scores of the images with different perception qualities through the perception quality evaluation indexes; Training the sorting estimation network according to the calculated perception quality evaluation score; Calculating the sorting content loss of the image generated by the generator of the generative adversarial network according to the trained sorting estimationnetwork, and guiding the generator of the generative adversarial network to train according to the sorting content loss; Carrying out super-resolution image reconstruction on the low-resolution imageaccording to the trained generator. The better image perception quality is obtained by directly optimizing the perception quality evaluation indexes by using the sorting estimation network, expansioncan be flexibly carried out according to requirements, and a generator can be restrained to generate super-resolution reconstructed images with different characteristics.

Description

technical field [0001] The present application belongs to the field of image processing, and in particular relates to a super-resolution image reconstruction method, device and equipment. Background technique [0002] Super-resolution image reconstruction technology has important academic research and industrial application value in the field of computer vision and image processing. The purpose of super-resolution image reconstruction technology is to reconstruct the corresponding high-resolution image based on a given low-resolution image, and to obtain as good a visual effect as possible and a small reconstruction error. [0003] In the reconstruction of low-resolution images, the learning method of convolutional neural network and the method of confrontation generation network are generally included at present. Among them, the learning method of the convolutional neural network uses the mean square error (Mean Square Error, MSE) to find the error between the high-resolut...

AI Technical Summary

Problems solved by technology

Compared with the mean square error, although this method can get a better visual effect, compared with the real high-resolution image, the texture details are not realistic enough.

On the other hand, the perceptual quality evaluation index uses feature extraction and further transformation operations, so it is non-differentiable, and gradient solving and backpropagation cannot be performed, which makes it difficult to optimize the perceptual quality evaluation index, and the visual Effect improvement is limited

Images