An Adaptive Image Registration Method Based on SURF and Least Square Matching

  • 賴 秉鈞

Student thesis: Master's Thesis


Image registration geometrically aligns two images and it is a pre-process for many applications such as image fusion change detection and image stitching In this thesis an efficient image registration method is proposed to register two images which are taken at different viewpoints and correct the misalignment caused by local geometric distortion In order to register multiview images a 2D projective transformation is used to describe the relation between two coordinates A feature-based method is used to estimate the global transform model SURF is used to speed up and save memory and the modified RANSAC is proposed to quickly and accurately estimate transformation parameters Furthermore the adaptive partition framework is designed to correct local distortion It divides image to properly blocks and least square matching is used to refine local parameters for each block Experimental results show that the proposed method has high accuracy and can deal with local misalignment caused by geometric distortion
Date of Award2017 Sep 14
Original languageEnglish
SupervisorShen-Chuan Tai (Supervisor)

Cite this