A Robust Feature-Based Registration Methodfor Differently Exposed Image Sequences

• Main Authors: Yen-liang Chen, 陳彥良
• Other Authors: 吳俊霖
• Format: Others
• Language: zh-TW
• Published: 2007
• Online Access: http://ndltd.ncl.edu.tw/handle/92753043115389784294

碩士 === 國立中興大學 === 資訊科學系所 === 95 === In feature-based image registration method, the most challenge is to find the features which have high matching accuracy between images. If the pairs of feature can be found correctly between images, the mapping function is constructed. It should transform one image to overlay it over the other one. Scale Invariant Feature Transform (SIFT) is a robust feature transform for feature detection and descriptor. The features are invariant to image scale and rotation, and are shown to provide robust matching across a substantial range of affine distortion, change in viewpoint, addition of noise and change in illumination. But if the source images have large change in illumination and the existence of similar objects, the matching will fail. In matching time, SIFT spends too much time on sorting. In this thesis, we use the median threshold bitmap (MTB) to solve the large change in illumination problem and estimate the texture parameter with normalized (R) and entropy (E) to improve the existence of similar objects. In feature matching, we use partial sum to improve matching time. The experimental results show that the proposed method can improve and solve the problems above effectively and shorten the matching time.