Study On Image Registration Algorithm With High Registration Ratio And Quick Velocity

Image registration is one of the key technologies applied in computer vision, object recognition, image reconstruction and robot vision, etc. It is also one of the basic tasks in image processing. It enables two or more images taken in different time, perspectives, shooting conditions and by different sensors, to be matched. Image registration has already been widely used in military, remote sensing, medicine, remote sensing image processing and other fields.Currently, image registration study is mainly divided into three categories: gray-based registration methods; registration methods based on understanding and interpretation of the image and feature-based registration methods. With the development of related technologies, image registration requires achieving high precision and accretion registration within a certain time.The thesis focuses on feature-based registration methods research, specifically as follows:①Firstly the paper illustrates the development of image registration techniques, background, introduces in detail feature-based image registration method, and analyzes its research status. And then the paper focuses on feature detection, feature description, feature matching problems and their realization of the importance of the whole registration process.②In order to effectively improve the problems in the feature detection technique, the paper recommends a multi-scale CSS feature detection algorithm, based on the analysis of the existing Harris, FAST, DOG feature detection algorithm and further improvement of the original CSS feature detection algorithm, which can efficiently detect the feature points of the image, improve processing speed, and characterize the appearance features of the images.③For the problems existing in feature-based image registration algorithm, multi-scale algorithm based on CSS multi-scale algorithm combines Daisy method of image registration algorithm and achieves highly efficient and accurate image registration.④After analyzing the advantages and disadvantages between multi-scale CSS image registration algorithm and DAISY image registration algorithm, it is found out that it is not a good method to solve the time effect of image registration algorithm by improving features detection and characterization. On the basis of the existing problem that matching algorithm is unable to solve high dimension data retrieval, here the paper proposes an improved algorithm (ELSH) combined with the traditional SIFT algorithm to realize fast and accurate registration between images.Through plenty of experiments and analyses, the paper verifies that multi-scale CSS feature detection algorithm with a combination of DAISY feature characterization registration algorithm can achieve precise and accurate images registration. After comparison, image registration algorithm based on ELSH and traditional BBF and LSH algorithm, can greatly reduced computational time and achieve fast and accurate registration.