Infrared And Low-Light Visible Image Fusion Technology Based On Multi-Scale Segmentation And Detail Enhancement

Image fusion is an image-enhancement technique that aims to fuse multiple images of the same scene to produce a more informative image for human eye decision making.Infrared and low-light image fusion is an image fusion method that is widely used and has high research value.Due to the good complementary characteristics of low-light images and infrared images,by effectively fusing these two images,both the infrared target information can be accurately detected and the richer background texture information can be obtained.In the last decade,multiscale algorithms have been the dominant approach to image fusion problems.The basic idea of this approach is to decompose the image into several sub-layers,each of which has unique features.Then,according to the characteristics of each sub-layer,different fusion rules are designed for processing,and finally all sub-layers are combined into a complete image according to the fusion rules.However,the traditional multi-scale algorithm also has certain limitations and defects,for example,the ability to capture image edge information and detailed texture information is weak,and some information of the image is easily lost during the image segmentation process,resulting in problems such as blurred background texture and unremarkable infrared target of the fused image.In order to pursue higher image fusion quality,the following research works are conducted in this paper to address the above problems:1.To address the problem of significant loss of background information and texture features in the multi-scale decomposition operation of images,a method of Quaternion Non-Subsampled Contourlet Transform(QNSCT)is proposed.This method uses quaternion wavelets instead of the pyramid structure in the Non-Subsampled Contourlet Transform(NSCT),which can more effectively implement multi-scale decomposition operations in the NSCT domain.Because there is no down sampling operation in the process of decomposition and reconstruction,it has translation invariance and high computational efficiency.2.A method is proposed to refine weights based on guided filter(GF).By estimating the weights of the initial image and refining them using an improved GF,it can effectively suppress the appearance of noise and artifacts while highlighting the saliency of infrared targets,thus improving the fusion quality.3.In this paper,an image fusion framework based on guided filtering and quaternion non-subsampled contourlet transform is proposed.Weighting rules are designed for fusing sub-bands of different scales,with weighting rules for infrared saliency maps and intensity detail layers,and weighting rules for low-light feature layers and gradient detail layers.Before fusion,the detail information and target brightness information of the images are enhanced,making the fused image more consistent with human visual characteristics.4.In this paper,based on infrared and low-light image fusion technology,the fusion image is applied to the target detection research.Infrared and low-light image detection data sets are trained on different networks.Finally,target detection experiments are performed on the fused images using different algorithms,and the detection rates of singleand dual-modal images are analyzed.The detection performance of each fused image is also quantitatively analyzed.The experimental results show that the proposed algorithm not only extracts important visual information from the source image,but also preserves the texture information in the scene better.Meanwhile,the scheme outperforms state-of-the-art methods in both subjective and objective evaluations.