Image restoration method based on gating and context attention mechanism

Abstract

The invention discloses an image restoration method based on gating and a context attention mechanism. The method comprises the following steps: multiplying an adopted visible light data set image and a contour image thereof by a binary mask M, and obtaining a degraded missing image and an auxiliary information image respectively; constructing a data set containing the real image, the missing image and the auxiliary information image through the degraded missing image and the additional auxiliary information image; improving a context attention module by adding a receptive field module and a loss function, and constructing an improved generative adversarial network model based on gated convolution and a context attention mechanism; training a generative adversarial network model through the missing image and the auxiliary information image to obtain an optimal generative adversarial network model; and inputting a to-be-repaired image into the optimal generative adversarial network model to obtain an effective repaired image. Compared with a traditional algorithm, the method has obvious progress on the image restoration problem of large-area or irregular-area deficiency.

Description

Technical field [0001] The present invention belongs to the field of image repair technology, and is specifically related to an image repair method based on the gate control and context. Background technique [0002] Image as an important information carrier that perceives objective worlds, has the advantage that language text is unacceptable. The deletion of image information not only destroyed the visual presentation quality of the image, but also affects the follow-up application of the image. The image repair technology is predicted with the content of the damaged area, and maintains the natural transition between the damaged area and intact regional pixels, and reduces the status and information before the image is damaged as much as possible. Image repair has important practical significance, widely used in old photos repair, image rendering, biomedicine, etc. [0003] Traditional image repair method, such as a sequence-based image repair algorithm is possible on small area...

AI Technical Summary

Problems solved by technology

[0003] Traditional image inpainting methods, such as sequence-based image inpainting algorithms, are feasible in dealing with small-area repair problems such as local texture damage, but cannot be effective for large-area and global structural distortion problems due to the inability to capture the global information structure. repair

Images