Research On The Fusion Algorithm Of Infrared Image And Visible Image Based On Multi-scale Transform

Infrared and visible image fusion is an important part of image fusion and has attracted extensive attention from numerous researchers since the technology was introduced,making it a hot research area.To meet the needs of different scenes,image sensors for different purposes have been developed.Because a single sensor can only take one image,for example,infrared images do not have spectral information of visible images.Infrared imaging unit can capture thermal radiation information of hidden targets,the greater the amount of radiation of the object into the brighter the image,but it lacks the details,and has the advantage of all-weather work.Visible light sensor imaging principle is related to the intensity of light,the stronger the light,the clearer the image,with high resolution,texture and detail information.However,it is susceptible to environmental influences such as night.Image fusion technology provides a way to fuse infrared and visible images into an image with both rich texture details in the scene and prominent infrared targets.There is a wide range of prospects in the military,surveillance and others.In this paper,a fusion method based on non-subsampled contourlet(NSCT)is proposed.NCST has the advantages of shift invariance and multi-directional selectivity,but the multi-directionality of NSCT has limitations and is not truly infinite.Therefore,a NSST based method is further proposed,which has unlimited number of directions.The following lists the specific research work in this paper:(1)To address the problems of low clarity and contrast,indistinct targets in image fusion.A NSCT with saliency map and region-energy algorithm is proposed.Firstly,the improved Frequency-tuned(FT)method is used to obtain infrared image`s saliency map,then normalized to get saliency map weight;Visible image is enhanced by Single-scale retinex(SSR).Secondly,using NSCT decomposes images.New weights based on normalized saliency map and region energy fuse low frequency coefficients,which solve the problem of adaptive weighting of region energy that easily introduces noise;A modified "weighted laplace energy sum" is used to fuse high frequency coefficients.Finally,the fused image is derived by inverse NSCT.This method is compared with seven classical methods in six groups of images,and the results are optimal in information entropy,mutual information,average gradient and standard deviation,and the first group of images is suboptimal in spatial frequency.The fused images are rich in information,high definition,contrast and moderate brightness for human eyes,which confirmed its usefulness.(2)To improve the problem of unclear and poorly textures that exist in fused images.An algorithm based on global energy features and improved PCNN is presented.Firstly,the infrared image is dehazing by dark channel to enhance the clarity.Then source image is decomposed by NSST,and low frequency coefficients are fused using global energy features with modified spatial frequency adaptive weight;The texture energy is used as external input of PA-PCNN to fuse high frequency components.Finally,the outcome is obtained by inverse NSST.The results show the method is significantly ahead of the comparison algorithms in objective indexes.Subjectively,this method has higher sharpness and contrast and richer texture details.The multi-resolution color transfer(MRCT)algorithm converts grayscale images into pseudo-color images to further enhance the recognition and human eyes perception.