Research On DDRI Noise Reduction And Improved SURF Registration Of Aerial Remote Sensing Images

Image registration technology is a key research area in the field of image processing.Currently,this technology is widely used in fields such as aerial photography and remote sensing image processing,medical image processing,and object recognition and retrieval.Aerial photography and remote sensing images are prone to noise interference during the acquisition and transmission process,which can affect image quality and lead to inaccurate results and low accuracy in subsequent registration processing.Additionally,aerial photography and remote sensing images have large amounts of information,large fields of view,and complex ground textures.Traditional registration algorithms used directly for such image registration can detect tens of thousands of feature points,but the feature matching accuracy and time efficiency are low,and the registration effect is not ideal.Therefore,in-depth exploration of rapid registration for aerial photography and remote sensing images is of great significance.This paper focuses on the research of image denoising and image registration stages to address the problems existing in current registration algorithms for aerial photography and remote sensing image registration,and proposes the DDRI denoising model and OST-SURF fast registration method.The main research contents and achievements include:(1)This paper investigates denoising techniques for images prior to registration.Traditional denoising algorithms are ineffective in removing mixed noise and may result in blurred image edges.To address this issue,we propose the DDRI secondorder removal method for mixed noise in remote sensing images.Our method first builds a first-order denoising model using dilated convolution and Dropout Layer layers based on Dn CNN.Then,we adopt a nearest-neighbor pixel weighted median to replace the median of the original filtering window based on adaptive median filtering to perform a secondary process on the first-order denoising results.Experimental results show that the proposed method outperforms traditional methods in denoising mixed noise in remote sensing images,while preserving image edge details and texture features.These improvements provide a solid foundation for subsequent image registration.(2)This paper analyzes and investigates various image registration methods.We focus on the adaptability and performance differences of different registration methods on aerial remote sensing images.Specifically,we conduct registration experiments on five commonly used algorithms: SIFT,BRISK,KAZE,ORB,and SURF.We evaluate and compare their performance based on several indicators,including the number of detected feature points,feature point change rate,registration rate,matching accuracy,and algorithm running time.After comparing and evaluating these methods,we conclude that the SURF algorithm has relatively better overall performance,and we use it as the basis for improving and optimizing subsequent registration of aerial remote sensing images.(3)In the context of aerial remote sensing image registration,despite the good performance of the SURF algorithm,it still suffers from the problems of detecting a large number of useless feature points,high computational complexity for feature extraction,redundant feature descriptors,and insufficient real-time performance for registration.To address these issues,this paper proposes a novel OST-SURF fast registration method for aerial remote sensing images.Specifically,this method first calculates the feature point detection region based on the overlap ratio of aerial images,and then performs SURF feature point detection and extraction within the region.A 34-dimensional feature descriptor is constructed using a circular template with color information,and the feature matching is achieved by matching the feature descriptors.Finally,the MSAC algorithm is used to eliminate the mismatched feature point pairs to achieve accurate image registration.Experimental results demonstrate that the proposed registration method significantly improves the accuracy of feature matching and the efficiency of algorithm running time for aerial remote sensing image registration.