Multi-source Image Fusion Based On Super-resolution Reconstruction

In recent years,multi-source image fusion technology has been widely used in medical diagnosis,remote sensing monitoring,video surveillance,military and other fields.It can complement information in multiple images of the same scene captured by different sensors or the same sensor with different parameter settings.To get the image with more fully describes the scene information,which provides a more complete description about the observed object.However,the existing multi-source image fusion scheme cannot improve the spatial resolution of the original image during the implementation process,which limits the further application of the fusion result.The use of super-resolution(SR)reconstruction technology can reconstruct higherresolution images from low-resolution source images,laying the foundation for improving the spatial resolution of multi-source image fusion results without changing the imaging equipment.For improving the spatial resolution of the fused image,the existing two-stage method performs super-resolution reconstruction on the fused image or super-resolution reconstruction of the source image before fusion.However,in this step-by-step operation,the artifacts generated in the first step will inevitably propagate to the latter link,causing the final high-resolution results to be disturbed,causing the loss of source image information and the introduction of artificial information.To solve the shortcomings of step-by-step processing of image fusion and super-resolution reconstruction,this thesis proposes a multi-source image fusion algorithm based on super-resolution reconstruction.The main work of this thesis includes:(1)A new structure and texture component decomposition model to realize the learning of multi-component dictionary is designed in this thesis.To characterize the relationship between low-resolution images and their corresponding high-resolution images,the correlation between high-and low-resolution sparse coding coefficients is introduced into the model.(2)In order to improve the quality of multi-source image fusion,a new saliency measurement scheme is designed in this thesis to construct corresponding fusion results for image components with different morphological features.(3)In order to compensate the information loss during in the process of superresolution reconstruction,this thesis proposes a reconstruction residual compensation mechanism which can compensate the reconstruction residuals to the initial results of super-resolution reconstruction and fusion to improve the quality of the final processing results.Many experimental results show that the proposed multi-source image fusion algorithm based on super-resolution reconstruction in this thesis can retain the brightness and detail information of the original image better,and which is superior to other contrast methods in subjective and objective evaluation.