Image Fusion Method Based On Sparse Representation

In recent years,the sensor began to use in real life,the image information access becomes more simple.For different information contained in the same scene,different sensors will have different shooting modes.Thus making the focus of the information inconsistent.In order to improve the information,image fusion technology can be combined with different focus,so it is widely used.The successive lifting of remote sensing satellites leads to the acquisition of multi-spatial,multi-spectral and temporal resolution images of the same scene.However,due to the limitations of the sensor design itself,often in the case of high spectral resolution,can not get high spatial resolution.The emergence of image fusion technology to solve this problem,it can combine the advantages of two images together to obtain high spectral resolution and high spatial resolution.Compared with the existing remote sensing image fusion technology,remote sensing image fusion technology based on sparse representation has its strong adaptability,better preservation of structural information and spectral information and so on,which has aroused the concern of many scholars.In this paper,focus on the remote sensing image fusion algorithm,and propose a series of improvements for the fusion algorithm of low resolution multi-spectral image and high resolution panchromatic image.The goal of achieving the fusion effect is to ensure that the complete reservation of the spectral information is accompanied by a certain increase in spatial resolution.Specific work as follows:(1)The guided filter is based on the local linear model,select the appropriate guide image,the image value,brightness and color will be used as a template for the filtered image.Based on the similarity between the template and the template,the influence of the pixel on the fixed pixel is calculated,and the filter core is generated on the fixed pixel.In the early stage of the fusion,the use of guided filtering method can make the difference between the luminance map and the full color map become smaller,so as the way to pave for the later fusion effect.Sparse representation of the model into the image fusion framework to retain the structure of the image information and through the study of the relevance of the dictionary to deal with the same time can achieve the denoising effect.In this paper,the sparse fusion algorithm of remote sensing image based on guided filtering is proposed.The low resolution multi-spectral image and the high-resolution panchromatic image are selected as the fusion material,and finally the high-resolution multi-spectral image is merged.In the beginning,the IHS transform is applied to the multi-spectral image,and then the panchromatic image is used as the guide template to guide the brightness part of the IHS.According to the characteristics of the processed luminance and panchromatic image,adaptive dictionary training is carried out to obtain two different sparse representation coefficients.The sparse representation of the fusion rule with the larger image activity is the coefficient after the final fusion.(2)In the early stage,after the preliminary basis of the remote sensing fusion framework,the depth mining is carried out,and try to join the multi-scale transformation to the existing remote sensing fusion model.After the IHS decomposition and guidance filtering fitting process is complete,it is selected to put the shearlet transform into the whole frame,replacing the original image fusion,this time using the shearlet transform function in two images that need to be processed,respectively Separated into two subband coefficients:high frequency subband coefficients and low frequency subband coefficients.The low frequency subband coefficients are sparse by the sparse representation of the method of fusion,high frequency subband coefficients using regional equations and regional energy combined method of fusion.The improved fusion algorithm further improves the fusion effect.Better achieve the goal.