Local Property of Depth Information in 3D Images and Its Application in Feature Matching

In image registration or matching, the feature extracted by using traditional methods does not include depth information which may lead to a mismatch of keypoints. this paper, we prove that when camera moves, ratio difference keypoint and its neighbor pixel before after movement approximates constant. That means normalization is invariant movement. Based on property, all differences pixels constitute local depth-based feature, can be used as supplement feature. We combine with SIFT descriptor form new descriptor, experimental results show feasibility effectiveness descriptor.