Image restoration method based on multi-scale content attention mechanism, storage medium and terminal

Abstract

The invention discloses an image restoration method based on a multi-scale content attention mechanism, a storage medium and a terminal. The method comprises the following steps: inputting a to-be-restored image into a trained image restoration network for image restoration, wherein the image restoration network comprises an edge detection module, an edge generative adversarial network and a content generative adversarial network which are connected in sequence, the edge generative adversarial network comprises an edge generator G1, and the content generative adversarial network comprises a content generator G2, each of the edge generator G1 and the content generator G2 comprises a plurality of convolution layers, a plurality of residual blocks and a plurality of deconvolution layers, and a multi-scale content attention module is arranged between the deconvolution layers. According to the method, the features of the known patches are used as a convolution filter to process the pre-generated patches, and the consistency of the basic features and the overall style of the incomplete generation area and the surrounding background area is achieved through the multi-scale patches.

Description

technical field [0001] The invention relates to the field of image restoration, in particular to an image restoration method, storage medium and terminal based on a multi-scale content attention mechanism. Background technique [0002] The image completion algorithm adds special occlusion marks to the content that needs to be removed and processes it to generate realistic photos that meet the visual requirements. Image completion technology is a technology between image editing and image generation, which uses the domain image information and overall structure information of the damaged part to fill and repair the missing part. The repaired part and the original image need to meet the requirements of unified style, natural transition, and image semantics, and meet the visual needs of the observer, so that the observer cannot distinguish the real image from the generated image. [0003] For the task of image completion, researchers have proposed a variety of different method...

AI Technical Summary

Problems solved by technology

[0004] At present, the main defects of image inpainting algorithms are: (1) when the current image inpainting algorithm is missing a large area or a high texture area, the completion result does not conform to the image semantics; (2) the transition between the inpainting result and the original image is not natural enough , The style is not uniform enough, and the details and textures are blurred

Images