Generative adversarial network image restoration method based on multi-scale texture feature branches

Abstract

The invention discloses a generative adversarial network image restoration method based on multi-scale texture feature branches. The method comprises the following steps: constructing a to-be-restored image containing a missing region and a real image pair data set; constructing a generative adversarial network model based on multi-scale texture feature branches; training the generative adversarial network model based on the multi-scale texture feature branches to obtain an optimal generative adversarial network model; inputting a to-be-repaired image containing the missing region into the optimal generator network model, and outputting a final repaired result image. According to the method, a better repairing effect can be obtained for irregular random missing and large-area missing of the image, and the repaired image has better structure and texture features.

Description

technical field [0001] The invention belongs to the technical field of image processing, and in particular relates to a generative confrontation network image restoration method based on multi-scale texture feature branches. Background technique [0002] Image inpainting technology has always been a research hotspot in the field of image processing. For an image with a certain missing area, image inpainting technology can restore the information of the missing area by using the information of the known intact area, thereby repairing the image and maintaining Image structure and texture features. With the rapid development of this technology, it has been gradually applied in many fields such as cultural relics restoration, medical imaging, satellite remote sensing, etc. Therefore, it has high research significance and application value to carry out research on image restoration algorithms. [0003] In recent years, with the development of the field of computer vision, image ...

AI Technical Summary

Problems solved by technology

[0004] Image inpainting based on partial differential equations is often very blurry for images with complex textures or large missing areas; image inpainting based on texture synthesis performs well in texture approximation, but there are still some defects in natural image inpainting; When an image has both structural and texture components, image restoration based on partial differential equations is often combined with image restoration based on texture synthesis, that is, image restoration based on structure and texture is used, and the texture of the restored image is and structural characteristics can be well maintained

Images