Digital image restoration method based on structural information embedding and attention mechanism

Abstract

The invention discloses a digital image restoration method based on structural information embedding and an attention mechanism. The digital image restoration method comprises the following steps: firstly, carrying out size processing and data enhancement on a public data set image, and extracting structural information by adopting an RTV method; then, constructing an image restoration network based on structure information embedding and an attention mechanism, and simultaneously performing structure restoration and embedding guidance image feature reconstruction by utilizing a structure information and image feature sharing generator; using a hinge form structure information discriminator with spectrum normalization to discriminate the authenticity of the repaired smooth image; adopting a joint loss constraint model to optimize the direction; the trained model weight parameters and the constructed repair network can be used for completing a repair task and effect evaluation on the damaged image; according to the method, the structure information can be better represented, so that the repaired picture has a more reasonable structure and a more real visual effect, and subjective and objective test evaluation results and contrast experiments prove the effectiveness of the method.

Description

technical field [0001] The invention relates to the technical field of digital image restoration based on deep learning, in particular to a digital image restoration method based on structural information embedding and attention mechanism. Background technique [0002] The image inpainting technology infers the content of the missing area based on the known information of the damaged image, and makes the repaired image have a reasonable global semantic structure, fine texture details, and natural fusion of the boundaries of the missing area, and the comprehensive effect meets the requirements of human vision. The technology has a wide range of application value in the fields of image editing, cultural film and television special effects production, video cultural relics protection and military public security. Benefiting from the ability of neural networks to learn nonlinear complex mapping relationships between samples from a large number of training data sets and the power...

AI Technical Summary

Problems solved by technology

These methods prove that the attention mechanism can effectively capture global information and generate finer texture details, but the image block-level contextual attention mechanism will lead to invalid information extraction and similarity calculation when dealing with irregular mask restoration

[0005] Current image inpainting tasks tend to deal with complex semantic image inpainting problems with missing irregular regions. These inpainting processes have problems such as invalid feature extraction and training ambiguity, which cause artificial artifacts such as color differences, blurring, and obvious boundaries in the inpainted image.

Although some current methods use partial convolution, edge information, and attention mechanisms to improve the repair effect and generate interpretable structures and fine textures, there are also insufficient structural information for edge representations, unreasonable premise effects of sequence model architectures, and invalid feature extraction. And other issues

Images