Research On Infrared And Visible Light Image Fusion Algorithm Based On GAN

Image fusion refers to extracting and combining the most meaningful information from different source images to generate a more informative image that is beneficial to subsequent applications.The fusion of infrared and visible images is an important and frequently occurring problem in image fusion.Infrared images have obvious contrast,which can effectively highlight the target from the background even in bad weather.Visible images contain rich texture details,which are more in line with human visual perception.Infrared and visible image fusion combines these features to produce results with high contrast and rich texture information.Our paper proposes a fusion method of infrared images and visible images based on an improved generative adversarial network.The specific work is as follows:(1)This paper proposes a fusion method of infrared and visible images based on an improved generative adversarial network.By establishing a game confrontation between the generator and the discriminator,the generator can generate a fusion image that can deceive the discriminator,and the discriminator can complete the classification task of distinguishing between real images and fake images generated by the generator.The purpose of adding an attention mechanism to the generator is to construct information features by fusing the channel information in the local receptive field of each layer of the network,which helps the model pay more attention to the channel features with the largest amount of information and suppress the unimportant features in the channel.Therefore,the feature extraction capability of the network model is improved,and the deficiency of the general network architecture that cannot fully extract image features is made up.(2)In view of the image characteristics of infrared and visible images,this paper proposes a loss function based on SSIM as a part of the content loss function,which makes the network pay more attention to the brightness,contrast and structure of the image in the process of fusing the image.At the same time,the most suitable fusion weight is adjusted according to the experimental results,which not only fully retains the brightness information in the infrared image,but also balances the difference between the brightness and the gradient details in the fusion image,so as to obtain the optimal fusion result.(3)The above network model is designed by using the Pytorch platform.As an end-to-end model,the model avoids manually designing complicated activity level measurements and fusion rules like traditional methods.At the same time,the performance of the algorithm is verified through a large number of experiments.Using mutual information,information entropy,edge strength,spatial frequency,phase consistency,structural similarity and visual information assurance to objectively evaluate the quality of fused images,the subjective comparison and objective verification results show that the method proposed in this paper can finish better fusion of infrared images and visible images to obtain a more prominent fusion effect.