A face image inpainting method based on multi-column gated convolutional network

Abstract

The invention discloses a face image restoration method based on a multi-column gated convolution network, which obtains a face image to be restored and a mask, and inputs the face image to be restored into a generator with gated convolution Repair in the generated image to obtain the generated image; based on the perceptual loss method, use the implicit multivariate Markov random field to calculate the feature loss value between the generated image and the original image; calculate the confidence of the pixels in the generated image Assign value, and obtain a joint loss value according to the feature loss value and the two calculated loss values, which is composed of three parallel codec branches, each branch is set with different sizes of convolution kernels, and different faces are extracted respectively. The hierarchical semantic information improves the consistency of the global semantic structure; a gated convolution is incorporated to improve the repair ability of irregular missing areas; and the multi-scale neural block matching method is used to enhance the detailed texture of the face and improve the repair ability.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a face image restoration method based on a multi-column gated convolutional network. Background technique [0002] Image inpainting, also known as image completion, aims to infer the appropriate pixel information to repair the missing areas in the image. The core problem is to maintain the consistency of the global semantic structure and generate realistic images for the missing areas. texture details. Face inpainting, as a branch of image inpainting, is a challenging task to repair missing areas of faces. There are two main types of traditional image inpainting methods: one is based on texture synthesis. Such methods only use low-level pixel features, are difficult to capture the global structure and semantic information of the image, and cannot repair complex non-repetitive images like faces. The other is a search method based on an external database. After searchin...

AI Technical Summary

Problems solved by technology

However, these inpainting methods are limited by only transforming the image into a single-level feature space, ignoring the fact that inpainting involves different levels

Not only that, ordinary convolution uses the same convolution operation for all valid, invalid and mixed (missing border) pixels, and blurred repair results appear when repairing irregular missing areas, reducing the repair ability

Images