Research On SAR And Visible Light Image Registration And Fusion Methods

Multi-source image registration and fusion is an important means to achieve multi-source information integration and analysis,which has been widely used in remote sensing,medical,aerospace and computer vision fields.Due to the different imaging mechanisms of multi-source images,and there are usually large differences in gray distribution and inconsistent feature performance,which makes it difficult to extract the same features of multi-source images,therefore,it is difficult to achieve multi-source image registration based on gray or features.The contrast of traditional multi-source fusion images is low and targets in fusion images is not significant,which cause detrimental effect on visual analysis and computer processing.Therefore,new multi-source image stabilization registration and multi-source image enhancement fusion methods need to be explored to obtain more comprehensive and accurate multi-source image information,which will provide support for social development.This article focuses on the problems existing in the registration and fusion of radar and visible multi-source images,analyzes the difference characteristics of multi-source images,and carries out framework construction,edge extraction,stable registration,area division and enhanced fusion in two aspects of multi-source image registration and fusion through theoretical derivation,method research and experimental verification.The main research contents are as follows:1.The theoretical methods and applications of image registration and fusion are studied.In order to provide a theoretical basis for proposed stable registration and enhanced fusion algorithms,a research framework is built based on the transformation domain registration method,multi-scale decomposition method and large difference characteristics of multi-source images.2 Aiming at the problem of inconsistent features caused by the different gray of multi-source images,a new edge feature extraction method was proposed.Through single-line edge acquisition,edge image segmentation,and threshold filtering of multi-source images,the stable and consistent features can be extracted under the gray-level differences of multi-source images,which provides stable matching factors for multi-source image registration.3.Aiming at the problem of frequency leakage in traditional Fourier-Mellin transform registration,which leads to the false correlation peaks in multi-source image registration,a spectral leakage suppression method by window function filtering is proposed to obtain correct correlation peaks and achieve stable registration of multi-source images.4.A method of regional target background separation is proposed.The multiresolution algorithm is used to obtain the characteristic sub-graph in multi-scale and multi-direction.The characteristic sub-graph is used to calculate the area variance and threshold division of the characteristic sub-graph.All this work will lay the foundation for enhanced fusion.5.Aiming at the shortcomings of traditional pixel fusion image,such as low sharpness and contrast,and not prominent targets,a region-guided multi-source image fusion algorithm is proposed.The multi-scale and multi-resolution algorithm is used to accurately capture the image feature information in multi-scale and multi-direction,and the target features of region division are integrated into the multi-scale pixel fusion to achieve the information enhancement of the target region of the multi-source fusion image.The above methods solved the main problems between radar and visible light images registration and fusion through simulation experiments,and realized stable registration and enhanced fusion of multi-source images.