Study On Infrared And Visible Image Fusion Method

Infrared and visible light image detection has been widely used in various industries.These two types of images have very good complementary features.The former can overcome the harsh environment and distinguish the infrared target and background information in the scene,the latter has high resolution and rich visual effects.The fusion of infrared and visible images can provide more abundant and accurate information in the scene.Therefore,the research on the fusion of infrared and visible images has important theoretical and practical value in the fields of video security monitoring,night-time assisted driving,target detection and recognition.This paper first introduces the theory of image fusion technology,and analyzes three mainstream fusion methods in the field of infrared and visible image fusion.Based on this,a fusion method combining multi-scale decomposition and visual saliency is proposed.The multi-scale decomposition on these two types of images is implemented by the guided filter to obtain their base layer images and detail layer images,different strategies are used in different scale detail layers fusion and a method based on saliency is applied to the base layer fusion.The final result effectively preserves the texture structure and radiation information of the images.Aiming at the problems of complex design,difficult implementation or large computational complexity of most existing fusion algorithms,this paper proposes an improved infrared and visible image fusion model based on GAN network.With the input of source image pairs,the trained model directly outputs the predicted fusion image results,which performs higher fusion efficiency.In the model design process,the structural similarity of the image evaluation index is introduced into the network objective function design,which improves the overall effect of the fused image result in terms of contrast and brightness.The experimental results verify that the proposed model integrates the advantages of visible and infrared images perfectly,and the output enriches the image information,and has certain advantages compared with other methods.