Research On Multi-Sensor Image Fusion Based On Multi-Scale Transform Decomposition

The rapid development of sensing technology has further increased the need of people in the field of image fusion.Different sensors convey different information in different scenarios.The information expressed by a single sensor is always limited.Therefore,it is of great significance to obtain multi-sensor image information from different scenes and merge them to generate fused images with rich and comprehensive information which can help people further understand complex scenes.Based on the characteristics of multi-sensor images,we design the multi-sensor image fusion model based on multi-modal medical images and visible infrared images.The multi-sensor fusion model is verified the quality from both subjective and objective evaluations.1.Multi-modal medical image fusionMulti-modal medical imaging technology plays an irreplaceable role in clinical diagnosis.Computed Tomography(CT)images are widely used to locate dense structures.Magnetic Resonance Imaging(MRI)images are usually used to provide high-resolution anatomical information of soft tissues.Combining the two imaging technologies can further improve the applicability of clinical diagnosis.However,traditional multi-modal medical image fusion algorithms fail to maintain soft tissue and bones intensity information at the same time.In this paper,source images are decomposed into multi-scale levels by applying L0 gradient minimization.At the same time,the local energy weight map is constructed via the usage of the Gaussian filter and the local energy algorithm.Then,the saliency detection rule based on the highfrequency texture is used to detect the fine-scale details.The proposed algorithm completely retains the bones intensity and the tissue details,which shows superior performance in whether subjective and objective evaluations.2.Visible and infrared image fusionThe visible and infrared imaging technology is widely used in remote sensing and military detection.This is because the natural information transmitted by visible and infrared images is very complicated.They provide image information from opposite sides,which makes up for the deficiencies of each other to generate information with high robustness.We propose a visible and infrared image fusion model based on twoscale transform.In this algorithm,an average filter is initially used to decompose source images into a low-frequency sub-band retaining fine-scale features in small details and a high-frequency sub-band preserving high-contrast variation in intensity.A saliency detection fusion rule based on average filter and median filter is applied on highfrequency fusion to generate saliency maps.After obtaining the saliency maps,the strategy of Weighted Least Square optimization(WLS)is used to supplement visual details and reduce redundancy information to get the final fused detail layer.Meanwhile,we propose a low-frequency fusion rule,a saliency detection algorithm based on infrared thermal information enhancement and visible background detail conservation,which has good performance on maintaining brightness in infrared images and is able to capture background information in visible images.The main contributions can be summarized as follows :(1)We propose a multi-scale decomposition framework based on L0 gradient minimization,which solves the problem of incomplete scale separation and loss of detail texture.(2)We propose a contrast preservation method based on local energy and Gaussian filtering to better locate information with higher local energy.(3)A curve edge detection method based on gradient difference and the recursive filter is proposed,which can accurately capture gradient changes.(4)An infrared thermal area extraction algorithm based on saliency detection and information entropy supplementation is proposed to accurately capture infrared brightness changes.(5)A high-frequency fusion rule based on WLS optimization and saliency detection is proposed to supplement visual details and reduce infrared redundancy.