Research On Infrared And Visible Image Fusion Algorithm Based On Improved Generative Adversarial Network

In daily life,images are an important way for people to convey information.A single type of image contains less information,and cannot meet people's needs for various information.Through the fusion of different types of images,more abundant target information can obtained.At present,the fusion of infrared image and visible image is a hot research issue in the field of image fusion.Because the infrared image and visible image is through two different sensor devices to collect images,they are restricted by imaging principle,texture edge features and related factors,so the two different types of image fusion is more difficult,and need to design a specific image fusion method to solve this problem.In recent years,researchers have proposed a large number of image fusion methods,which divided into three categories: transform domain,spatial domain and neural network.However,traditional image fusion algorithms need to design complex activity levels and fusion rules manually,and obtained fusion images affected by subjective factors,so the image quality is easy to appear unsatisfactory.In addition,with the continuous improvement of hardware level,neural network-based methods applied in the field of computer vision and image processing.However,the neural network image fusion method proposed in recent years is not perfect in the use of the network,it will cause the loss of details in the training process,and it is difficult to obtain a better quality of fusion image.Aiming at the above problems,this paper studies the image fusion algorithm based on generative adversarial network,which divided into the following two aspects:(1)Aiming at the problems of traditional fusion algorithms that need to design complex fusion rules and poor fusion of image edge information,this paper proposes an infrared and visible image fusion algorithm that generative adversarial residual network.Firstly,in order to extract the depth features of the source image,the convolutional neural network structure based on residual module used in the generator.Secondly,the convolutional neural network combined with the convolutional layer and the full connection layer used as the network structure of the discriminator.The source image stitched into the generator to get the generated image.Finally,the discriminant used to distinguish the generated image from the target image,so that the generated fusion image is closer to the target image and has more detailed information.(2)To solve the problem that existing fusion algorithms only focus on preserving the details of a single source image while ignoring the features of another source image,in order to effectively preserve the details of infrared image and visible image,this paper proposes an image fusion algorithm based on Laplacian pyramid and generated adversarial network(referred to as Laplacian-GAN).Firstly,the guide filter used to decompose the source image to obtain the base layer and detail layer of the image.Secondly,Laplace pyramid transform used to fuse the base part,and generative adversarial network used to fuse the detail part.Finally,the fusion base image and detail layer image weighted to obtain the final fusion image.The fusion image effectively retains the infrared features in the infrared image and the detail information in the visible image.The experimental results show that the two fusion algorithms proposed in this paper can effectively retain the feature information of the source image and significantly enhance the quality of the fusion image.The comparison of these two algorithms with other multi-scale fusion methods and neural network fusion methods shows that the proposed algorithm not only obtains better visual perception subjectively,but also performs better in objective evaluation indexes.That is,compared with other algorithms in the six indicators,the average increase of 9.29% and 10.21% respectively,thus verifying the effectiveness and feasibility of the proposed algorithm...