An Image Super-Resolution Reconstruction Method Based on Multi-scale Generative Adversarial Networks

Abstract

The invention discloses an image super-resolution reconstruction method based on a multi-scale generative confrontation network, which uses a multi-scale SENet module as a generator to finally form a multi-scale generative confrontation network; then determines the target loss function to complete the pre-training process of the generator , improve the convergence speed of the multi-scale generation confrontation network; through the establishment of the multi-scale generation confrontation network, the feature extraction of the high-frequency information of the LR input image is realized, and then the HR image is reconstructed with the input result of the bicubic interpolation, and the discriminator To identify the authenticity of the reconstructed input, and weight the reconstructed mean square error loss and confrontation loss as the final adjustment objective function. After adjusting the output HR image through the objective function, the entire image reconstruction process is completed. The invention can extract more high-frequency information details of the LR input image, so that the HR image with better display effect can be generated after image super-resolution reconstruction.

Description

technical field [0001] The invention relates to an image super-resolution reconstruction method based on a multi-scale generation confrontation network. Background technique [0002] The image super-resolution (Single image super-resolution, SISR) reconstruction task refers to learning more image details from a low-resolution (Low-resolution, LR) image input to generate a high-resolution (High-resolution, HR) image. Since HR images can learn important details of the original images, SISR technology is widely used, including video surveillance, medical diagnosis, face recognition, etc. Traditional image super-resolution methods are mainly interpolation algorithms, such as bicubic interpolation and nearest neighbor interpolation, which estimate the value of unknown pixels in SR images by using fixed or structure-adapted kernel functions. Although the interpolation algorithm is more efficient, the result obtained is blurred and the edge details are smoother. [0003] In recen...

AI Technical Summary

Problems solved by technology

[0005] However, the above-mentioned methods are poor in extracting high-frequency information from low-resolution input images, so that the display effect of high-frequency information details on low-resolution input images is not good after image super-resolution reconstruction.

Images