Voting-Based Matching for Near-Planar Objects with Drastic Illumination Variation
Haiyong Xie, Jingqi Yan · 2010
Image registration is a challenging task when illumination changes drastically. In this paper, we propose a novel algorithm, APCurve, to address this problem. APCurve matches all principle curves extracted from the sensed image and reference image based on voting matrix. All principal curves are extracted because they are robust against illumination change. Voting matrix is a simple method to match all principal curves. We test APCurve on a set of images taken under drastically changed illumination condition, and experimental results show it is effective and fast. An experimental comparison between APCurve and SURF is presented, which shows APCurve outperforms SURF when the object is textureless. Finally, the successful application of APCurve in highlight removal task proves that our approach is useful in real practice.