Research On Thermal Infrared And Visible Image Fusion Method

Due to the different physical characteristics and imaging mechanisms of thermal infrared and visible images,their information can complement each other.The fusion of infrared and visible images can improve the accuracy and reliability of subsequent tasks such as target detection and recognition.Thermal infrared images can provide thermal information of the main target,while visible images can provide texture,shape,and color information of the image.By fusing these two images,more comprehensive image target information can be obtained.In military,security,and medical applications,thermal infrared and visible image fusion has been widely applied.This paper proposes a fusion method for sequence images and a fusion method for infrared and visible images based on generative adversarial networks,targeting the characteristics of sequence images and the shortcomings of existing image fusion algorithms.The main research work is as follows:In response to the highly correlated background information between image frames in thermal infrared and visible sequence images,which is not considered in traditional image fusion methods.In order to better fuse sequence images,this paper proposes a fusion method for infrared and visible sequence images based on joint low-rank sparse decomposition.The specific implementation steps of this method are as follows:firstly,the source sequence image is decomposed into three parts through joint low-rank sparse decomposition.The first part is the common low-rank component,the seecond part is the specific low-rank component,and the last part is the specific sparse component.Secondly,a weighted fusion strategy composed of phase consistency and guided filtering is used for the fusion of specific low-rank componets,while an average weighted fusion strategy is used for the fusion of specific sparse components.Finally,the common low-rank components and the fused two components are reconstructed to obtain the final result.This article selected Nato_Camp,Bristol Eden Project,and OSU Color and Thermal,three commonly used infrared and visible image sequence datasets,selected seven fusion algorithms for subjective and objective testing.The experiment shows that the method proposed in this paper has certain advantages.An infrared and visible image fusion method based on generative adversarial networks is proposed to address the impact of manually designed rules on traditional fusion methods and the inability to extract more reserved information from the source image.The specific implementation steps of this method are as follows:firstly,multi-scale decomposition is performed on the source image to obtain detail layer and base layer.Secondly,design the network structure of the generator and discriminator.The generator adopts a dual-branch network architecture,with inputs being the detail layer of infrared and visible images and the basic layer of infrared and visible images.There are two discriminators,which are used to distinguish between the two source images and the fused image.Finally,the SGD optimizer and RMSProp optimizer are selected to update the discriminator and generator parameters respectively,in order to improve the efficiency of network training.In the experiment,31 pairs of infrared and visible image pairs were selected from the TNO dataset as the training set to train the network,and 5 pairs of source images were selected from the TNO dataset to test the network.Subjectively and objectively test the fusion results obtained by comparing them with the fusion results obtained by nine different methods.The experiment shows that the visual effect and evaluation indicators of this method are good.