Research On Multi-source Image Fusion Based On Multi-scale Transform And Sparse Representation

Image fusion is an integrated representation method for multi-source image which depends on specific processing means and fusion rules.It can realize more comprehensive and accurate expression of the observed object.The fused image has higher resolution and greater information gathering degree than the source image.The rich features of the fused image are conducive to the subsequent detection and identification of tracking targets.Therefore the fusion system reduces the uncertainty in the decision-making process which has considerable economic benefits.On the basis of analyzing the deficiencies of the existing multi-source image fusion method,the key steps on accurately locating and extracting clear region of the source image are focused in this paper.Research on multi-source image fusion algorithm in multi-scale transform and sparse representation domain is conducted,and the main research contents are as follows:1?A fusion algorithm for infrared and visible images based on saliency analysis and non-subsampled Shearlet transform is proposed for highlighting the infrared target in the fused image.Firstly,the contour of the infrared target is extracted by using the saliency analysis method based on super-pixel segmentation.And then the multi-directional edge detection operator is used to accurately positioning the boundary of the target region.Finally,morphological method and adaptive thresholding method are usded to remove the false focus region and identify fusion decision map.As the result shown,the proposed method can locate the target region of the infrared image accurately.And the fused image can not only highlights the target region,but also retains the rich details of the background.2?For the problem of selecting clarity measure and improving the positioning accuracy of the focused region,a multi-focus image fusion algorithm based on focused region extraction is proposed.The saliency analysis method is used to locate the boundary of the focused region and eliminate the false contour of the focused region.To remove the pseudo-focused region,spatial frequency map of saliency analysis result is combined with watershed algorithm,which achieves precise segmentation for the focused region.The proposed method has improved the clarity of the fused image indeed and suppressed artifact interference preferably.3?An infrared and visible image fusion algorithm based on target separation and sparse representation is proposed to solve the problem of severe noise interference.The infrared target region is exactly separated by using regional statistical characteristics and kernel density estimation clustering method.On this basis,KSVD is used to get the sparse representation of the background.By this way,noise suppression capability of the method is enhanced,and fusion effects are superior to other fusion methods.4?It is easily to introduce artificial side effects which leads to blurred edges or artifacts appear in the fused image by traditional image fusion methods based on multi-resolution analysis.To solve the problem,a fusion method based on sparse decomposition is proposed.Robust Principal Component Analysis(RPCA)method is used to extract the contour information of the focused region.Then using the guided filter to enhance the edges and locate the focused region exactly by the spatial frequency map of the enhanced image.Experimental results show that the proposed fusion methods have obvious advantages,and the fusion effect has been significantly improved.Subjective and objective evaluations both show that the performance of the proposed fusion methods is better than the classical fusion methods.