Image restoration method and system based on generative adversarial network and application thereof

Abstract

The invention discloses an image restoration method based on a generative adversarial network and a system and application thereof, wherein the method comprises the steps of obtaining an optimized generated defect image boundary map through a trained boundary generation model in a first generative adversarial network; in a second generative adversarial network, training a restoration model by taking the original complete image, the original defect image and the optimized generated defect image boundary map as input to obtain a trained restoration model; and performing image restoration throughthe trained boundary generation model and restoration model. The restoration method provided by the invention can accurately restore the image defect area, significantly inhibits the image generatedby the defect area from generating blurring or artifacts, and is especially suitable for medical image restoration with high requirements for restoration accuracy and the like.

Description

technical field [0001] The invention relates to the technical field of image restoration by generating an adversarial network. Background technique [0002] With the wide application of deep learning methods based on convolutional neural network (CNN) in the field of image repair, especially the technology of generative confrontation network, a large number of related methods and models have been used to solve the problem of natural image repair, and further Applied to the restoration of medical images, in most cases, it can achieve better restoration results in different situations, but there are still many defects to be overcome. [0003] For example, in a prior art solution, image restoration is realized through a content generation network and a texture generation network, wherein the content generation network is directly used to generate an image, and the possible content of the missing part of the image can be inferred; the texture generation network is used to enhanc...

AI Technical Summary

Problems solved by technology

However, there are certain defects in the priority calculation method. For example, after many repairs, the data item drops sharply or even becomes zero. At this time, because the product operation is used in the calculation, even if the confidence item value is very high, but If the calculated priority value is not high, the obtained priority value is no longer accurate, and it is easy to cause the low-texture area to lag and repair the high-texture area to expand excessively, and the lack of structural information in the priority calculation will also lead to its Inability to repair edge regions preferentially, resulting in structural breaks in the repaired image

[0017] Some other existing technical solutions have similar defects. In the part of the repaired image corresponding to the defect of the source image, structural breaks, information loss, or blurring or artifacts often occur, and a clear and complete repaired image cannot be obtained.

[0018] In natural inpainting images, the above defects can be ignored to a certain extent, but in medical image inpainting, the above defects will greatly affect medical diagnosis, so they must be overcome

Images