Illumination-Robust Image Matching Based on Non-linear Correlation

Changes in lighting are unavoidable in natural scenes and they have a big effect on the way an object looks, awhich makes it a common challenge to match images taken under different lighting conditions．This paper presents an image matching method which is robust to varying illumination． Discriminative image pixels are extracted by segmenting the original image, awhich are then described by two parts, aone is about the status of region changes and the other is about the change degree in a region．In the matching stage, athe angle of feature vectors is used as the similarity measure in the high dimensional space directly．This similarity measure is invariant to the linear transformation of feature vectors．Compared with SIFT and normalized correlation, athe proposed method needs no primary features such as edges, acorners, agradient or shapes which are very unstable in low contrast and low NSR images．Experimental results demonstrate the validity of our method．