Research On The Registration Of Short Ware Infrared And Visible Near-infrared Image

Since 21th century,our hyperspectral imaging technology has developed rapidly,which makes it lead the world as our science research following the time tide.The modern hyperspectral imager can be applied for different bands,these are visible near-infrared,the short wave infrared,themiddle wave infrared and the long wave thermal infrared.So the hyperspectral imager application scenarios can be more complicated,meanwhile,the accuracy of remote sensing image processing is highly demanded.The image registration is an important step in remote sensing image processing,and is essential to the subsequent processing.The image registration is the process of matching and superimposing two or more images obtained from different variable conditions,including time,space,sensor or weather conditions.In the field of remote sensing imaging,a well registration between the images is an indispensable step in subsequent image processing.At present there are many algorithms of image registration,but exist lots of remote sensing image registration methods,and most of the present mature registration methods are focus on different sensors with the same resolution.The results of remote sensing image registration with multivariate sensors and multiple resolution are very little,so this paper do this.The in depth study of this subject has important significance to the development of the hyperspectral imager and remote sensing image processing.In this paper,the study focus on the remote sensing image with multivariate sensors and multiple resolution,use the short wave infrared and visible near-infrared image by themultispectral and multi-mode hyperspectral imager as the research object.First,we use the current image registration algorithm to do experiments and then make the image gray-scale registration to do experiments,but the results were not ideal.After a new algorithm searching for the homologue points based on edge features and is proposed in this paper,the result of the image registration is obtained and validated.Relative to the current image registration algorithm,the algorithm improve the accuracy of remote sensing image registration with multivariate sensors and multiple resolution in this paper.The algorithm idea of this paper:edging detection after gray-scale target image and reference image,getting edge features and filter out useful features,and then extracting connected domains,geometric figure fitting of straight line,curve and circle in connected domains,and getting the homologue points,validating and then filtering the correct homologue points,obtaining the image transformation relation with the preserved homologue points,finally getting the matching results comparison image.