Image fusion method and system based on generative adversarial network, and storage medium

Abstract

The invention discloses an image fusion method and system based on a generative adversarial network, and a storage medium. The method comprises the steps that a sample set is used for pre-training a fuzzy region recognition model, and a mask image of a fuzzy region marked by each image sample in the sample set is output, wherein the sample set comprises the image samples and a fusion image label Ir; multi-channel images formed by stacking the image samples and the corresponding mask images are input into a fusion model to be trained, wherein the fusion model comprises a generator and a discriminator; a fusion image If output by the generator and a fusion image label Ir are input into the discriminator to be subjected to adversarial training; and a to-be-fused image is input into the trained fuzzy region recognition model and the fusion model to generate a fused image. According to the invention, image fusion can be realized only by collecting a few multi-focus images, the time cost andhardware cost of image fusion can be effectively reduced, and the method is especially suitable for fusion of super-large pathological section images.

Description

technical field [0001] The invention belongs to the technical field of image processing, and more specifically relates to an image fusion method, system and storage medium based on a generative confrontation network. Background technique [0002] The purpose of image fusion is to fuse multiple input images into one fusion image. Compared with any single input image, the fusion image can provide more information for human or machine perception, including more information and clearer images. Fusion images are beneficial to the further processing of cervical cancer cell pathological sections. At present, under the optical microscope imaging system of cervical cancer cell pathological sections, by setting the focal length of the optical lens, only objects within the depth of field are clearly visible in the image, and other places are generally blurred. Therefore, in order to obtain a larger depth of field The image, usually adjust the focal length to shoot multiple images for ...

AI Technical Summary

Problems solved by technology

[0003] Traditional fusion algorithms based on transform domain and space domain can obtain high-quality fused images, but they still may lose some information of input images due to inaccurate fusion decision mapping

In recent years, with the large-scale application of convolutional neural networks in the field of image processing, more and more people have used convolutional neural networks in the field of image fusion, developed image fusion algorithms based on deep learning, and improved the effect of fusion. However, limited by the size of pathological slice images (on the order of billions to tens of billions of pixels), using traditional image fusion algorithms or existing fusion algorithms based on deep learning to fuse multi-focus slice images requires hardware cost and time efficiency. exist challenges

Images