Research On Adaptive Features For Large-scale High-resolution Remote Sensing Image Registration Method

With the development of remote sensing technology,the image width and resolution have been continuously increased.More and more high-resolution large-scale remote sensing images have been used in various fields,such as resource and environmental monitoring,land and resource planning,military and national defense,etc..Using image registration technology to register and align these images is the basis for studying remote sensing of large scale images.However,The multi-temporal large-scale high-resolution remote sensing images have more complex texture features,significant background changes,larger image widths,and a single global image transformation model cannot accurately describe the image transformation of local regions.It becomes more difficult to perform high-precision image registration.In addition,in order to obtain a wider range of image target areas in related research,in addition to high-precision registration of large-scale remote sensing images,it is also necessary to stitch these registered remote sensing images.The traditional methods usually require huge time consumption.Therefore,completing high-precision registration and fast stitching of large-scale high-resolution remote sensing images has become a research hotspot and difficulty.The main contributions of this thesis are as below:1)In view of the problems of large scene changes and difficulty in feature extraction and matching in local regions of large-scale remote sensing images,this paper proposes a registration method suitable for multi-temporal large-scale high-resolution remote sensing images.It is an image registration method of adaptive regional and multi-features(Adaptive Regional Multiple Features,ARMF).Since the multi-temporal large-scale remote sensing images are characterized by large background changes in the same area of the image,the number of feature extractions is insufficient and the feature matching is easy to fail during image feature detection and matching.Feature-based image registration search strategy,which uses an adaptive pyramid image region amplification strategy to find enough matched features.Then,we adaptively search and extract multiple types of features(gradient feature,phase feature,and line feature)in image local regions,which can represent large-scale image local regions more effectively.We use the feature matching error as a criterion for the start and end of image region amplification to adaptively select appropriate combined features as descriptors of image regions.Finally,the(As-Projective-As-Possible,APAP)is used to calculate the local area image transformation model to improve the registration accuracy of multi-temporal large-scale high-resolution remote sensing image.2)Aiming at the problems high time cost and improve the efficiency of image stitching in the search and extraction of local overlapping regions of images in large-scale remote sensing image stitching,Aiming at the problems of poor real-time performance and high time cost in searching and extracting partial overlapping regions of images in large-scale remote sensing image stitching,this paper proposes a fast remote sensing image stitching method based on APAP and local area mutual information similarity measure to improve the large-scale remote sensing image stitching speed.The method first uses the image annular boundary region extraction strategy,uses the mutual information similarity measure to quickly determine the image overlapping area,and uses the number of image feature matching and the correct feature matching rate to determine the image overlapping area again.If the matching fails,we extract the annular boundary area of the image for the second time and perform feature matching again until the accurate image overlap area is found.Finally,the image grid transformation and linear fusion in APAP are used to realize the stitching of the image overlap area.Our annular boundary image overlap region search strategy improves the speed of image stitching,reduces time consumption and ensures the accuracy of the extracted image overlap region.3)Aiming at the registration and stitching requirements of large-scale high-resolution remote sensing images,this paper designs and develops an auxiliary software for remote sensing image registration and stitching.The software includes the automatic remote sensing image registration function module of ARMF and the fast remote sensing image stitching function module,which integrates the above-mentioned remote sensing image registration algorithm and stitching algorithm,it also provides users with a convenient large-scale remote sensing image registration and stitching method.