Research On Image Fusion Algorithm Based On Nonseparable Shearlet Transform And Sparse Representation

The purpose of image fusion is to obtain the same scene image from the sensor of the same sensor under different conditions or different imaging mechanism to produce an image that conforms to computer processing or human observation.The fusion image is richer and more accurate than the source image information,which is beneficial to the subsequent detection,recognition and tracking work,and the image fusion technology has a good application prospect.The process of image fusion involves data processing means and fusion rules.Aiming at the lack of directional information,the lack of obvious features of image innovation and the decline of clarity in some parts of image fusion technology,the following explorations and studies are made in this paper on the image fusion algorithm of the discrete nonseparable shearlet transform and sparse representation:Firstly,a multi-scale decomposition tool for discrete nonseparable shearlet transform is studied,and discrete nonseparable shearlet transform is a direct shear operation using shear operators,which makes the discrete nonseparable shearlet transform define dispersion directly in the continuous domain,improves the overlap between shear supports,and optimizes the direction selection.Then an image fusion method based on discrete nonseparable shearlet transform and joint sparse representation is studied,which uses joint sparse representation method to fuse the discrete low-frequency after discrete nonseparable shearlet decomposition,and parameter adaptive pulse-coupled neural network fusion for high-frequency.the method solves the problem that the innovative characteristics are not obvious and does not conform to the visual observation.Finally,an image fusion method based on discrete nonseparable shearlet transform and convolution sparse representation is studied,which uses convolution sparse representation algorithm for the low-frequency subgeneration after discrete nonseparable shearlet decomposition,and then uses field feature fusion for high-frequency sub-generation,because the convolution sparse representation is sparse coding based on the whole image.To avoid the loss of image clarity measurement information,the problem of limited detail preservation and high sensitivity of registration errors is effectively solved.The experimental results show that the above fusion methods have obvious advantages,improve the clarity and detail preservation ability of the image after fusion,and the main objective evaluation results are optimal.