Remote Sensing Image Registration Algorithm Based On Pulse Coupled Neural Network

Image registration is an important task in image processing.It is the process of matching two or more images with different time,angle,and sensor to the same coordinate system.With the rapid development of the remote sensing technology,the spectral resolution or spatial resolution of remote sensing images from different sensors is usually different.The image registration with high precision of multi-source remote sensing images is the prerequisite for image fusion,image mosaic,target recognition,and change detection.However,in the multi-source remote sensing image registration,most current segmentation based image registration methods divide the segmentation process and the registration process into two independent steps.This will result in lacking feedback of the registration quality to the segmentation results.Moreover,the multi-source remote sensing images have large gray level differences.Using only one type of feature detection method may lead to lose other potential feature information and reduce the number of correct matches,which can affect the quality of image registration results.Therefore,to obtain the high precision registration results,the paper conducts research on multi-source remote sensing image registration algorithms based on pulse coupled neural network(PCNN)segmentation method and feature-based image registration methods.The main research content and innovation points of the paper is as follows:(1)In order to solve the problem of lacking feedback between the image segmentation and the image registration in traditional registration methods.The paper proposes a segmentation based adaptive PCNN registration method.Currently,most remote sensing segmentation based image registration methods divide the segmentation process and the registration process into two independent steps.This will result in lacking feedback of the registration quality to the segmentation results.The registration results will be badly influenced by only using a single segmentation process when dealing with different registration datasets.Therefore,the paper proposes a segmentation based remote sensing registration methods,which integrate segmentation and registration together.In the method,the image registration quality in each iteration will feed back to the pulse coupled neural network parameters by the slime mould algorithm.The PCNN segmentation parameters are adaptively optimized and re-segmented to obtain the best registration quality.(2)The paper proposes a PCNN and point features based remote sensing image registration algorithm to solve the problem of poor matching effect of single feature.The feature-based methods have good performance on remote sensing image registration,but only one type of feature detection method may lose other potential feature information and reduce the number of correct matches.It will affect the quality of image registration result for remote sensing images with complex content.Moreover,the number of correct matches will also affect the fitting precision of the geometric transformation matrix.Therefore,the paper focuses on improving the single feature registration and the number of correct matches.In the method,the UR-SIFT algorithm and adaptive PCNN segmentation method are used to obtain point and region features,respectively.The scale principal direction constraint strategy and the joint score nearest neighbor distance ratio are used to match point and region features,which will obtain high-precision multi-source remote sensing image registration result.The paper proposes two remote sensing image registration algorithms based on the adaptive PCNN network model.These algorithms have conducted the issues such as adaptive feedback between the segmentation process and registration effects,parameter adaptation of PCNN model,and matching strategies for segmentation regions.The effectiveness of the methods is verified through the comparative experiments of different datasets.The methods designed in the paper have some theoretical and practical significance for the multi-source remote sensing image registration.