The Research And Implementation Of Image Registration System Based On Feature Point Matching

Image Registration is a technique that relates or aligns different images taken from different viewpoints, at different times, or by different image sensing devices. Till now, image registration technology is widely used in the field of virtual reality(VR), video compression, image super-resolution, intelligent surveillance system, etc, which came to be an active research area in computer vision during recent years.Feature-based,direct pixel difference optimization based, and Fourier based method are three typical ways of image registration, which have its own appropriate application area separately. The key problem focus on how to increase the algorithm speed, to increase the registration precision and to enhance the robustness of these registration methods.This paper did research on overlayed image registration in the same scene.The main works are detailed as followed:①The meaning, methods, flow of feature-based image registration are disserted. The effect of ths critical process is analysed and compared.The current image registration methods are introduced and their advantages and disadvantages are analysed.The advantage of key points is analysed. SIFT algorithm is improved and used, which is of better results and less calculation. Compare with the original algorithm, the new algorithm increases real-time performance.②The meaning of fundamental matrix is introduced and the advantage of fundamental matrix using in the correspondence establishment is analysed. Some robust algorithm of estimating the fundamental are reviewed. A cost function related to fundamental matrix is introduced so that the threshold value problem in Euclidean distance could be solved.③This thesis did in-depth study in RANSAC algorithm, and proposed a weighted robust algorithm based on gray level to make the registration more efficiency.At the same time, a robust estimating method base on layered progressively is presented to employed in the initial parameter estimation when orientation of the images is great.④"Modularization"thought is used to designed the registration system.Plug-in unit type structure makes the system flexible and extendible.At last, experiments and analysis are given to show the new algorithms we proposed have a good stabilization, robustness and application performance.