Application Of Non-subsampled Contourlet Transform To Multi-source Image Fusion

With the increasing development of imaging technology,its application has becomes more and more extensive.Image has gradually become an important carrier for expressing information.On the one hand,currently the single-sensor imaging mode can only describe the scene from single angle,cannot obtain a complete expression of the target,which affects the effect of image data analysis.On the other hand,it is occupied when storing and reading the image data of a large amount of space is also very time consuming.Therefore,it is necessary to collect information from multi-modal images and integrate them into a single image so that richer information can be extracted from one image.The application of image fusion technology to image processing can meet this demand.Different sensors can describe images from different angles.Using image algorithms to extract image information from different sensors into an image is very helpful for computers to process image information.This paper studies pixel-level image fusion,using multi-scale analysis tool Non-subsampled Contourlet Transform(NSCT)to study the multi-source image fusion method,by analyzing the characteristics of the coefficients of NSCT transform,proposes a multi-source image fusion algorithm,mainly discussed Multi-focus image fusion method and infrared image fusion method.For multi-focus image fusion,in order to more comprehensively describe the definition of image pixels,improve the quality and texture of the fusion image,using the statistical features of information entropy,Laplace energy sum and average gradient,combined with edge detection technology,this paper designs a multi-focus image fusion method.Based on the transform domain fused in frame NSCT,the local information entropy and local Laplacian energy are used to calculate the pixel matching degree of the multi-focus image for low-frequency component fusion.For the high-frequency components,the coefficient correlation of the NSCT domain is used to calculate the sibling correlation weight,and the fusion rule is designed in combination with the average gradient in the domain window.Edge detection technology has high efficiency in image edge extraction,which can greatly improve the clarity and texture characteristics of the fused image.After the algorithm is initially fused,edge detection is performed on the high-frequency components,and the edge image is gradually overlaid on the preliminary fused image to produce the final fused image.Aiming at the fusion of infrared image and visible light image,in order to make the fusion image can obtain better definition and detailed target information.In this paper,the combination of the Laplacian energy sum weighted by Mahalanobis distance and the improved frequency domain coordination algorithm of guided filtering realizes the fusion of infrared image and visible light image.Visible light images are easily affected by light,resulting in low image contrast,and the histogram equalization of visible light images can improve its contrast clarity.The source image is preprocessed and decomposed by NSCT,and the low-frequency components in the transform domain are weighted and fused using the infrared image saliency map design weight.A distance-based weight calculation method is designed using Mahalanobis distance for high-frequency components to calculate the local Laplacian energy and perform large-scale fusion.After the components are processed separately,the fusion image is obtained through NSCT inverse transformation.