Image feature detection and matching is a key problem in many applications of computer vision, such as image registration, 3D reconstruction, object detection and video understanding. In this dissertation, the main work is focused on feature point detection and matching, and the main contributions include: 1． The exterior product energy is introduced by using exterior product operations of image gradients, which can effectively enhance corners and edge points with large curvature in images. Then, a feature point detector is proposed, namely maximum exterior product (MEP) detector. The MEP detector can not only detect corners and edge points with large curvature but also can detect blobs in images. In addition, it is invariant to image rotation and linear change of illumination. 2． Based on the inner and exterior products of image gradients, gradient correlation and feature vector field are introduced and two novel descriptors, gradient correlation descriptor (GCD) and feature vector descriptor (FVD), are constructed. Both the GCD and FVD are invariant to image rotation as well as to linear change of illumination. In addition, experimental results show that these two descriptors have also a good adaptability to image affine distortion, blurring, JPEG compression and nonlinear change of illumination. 3． Inspired by Harris corner detector, Harris correlation and Harris feature vector field are introduced and two novel descriptors, Harris correlation descriptor (HCD) and Harris feature vector descriptor (HFVD), are proposed. Both the HCD and HFVD are invariant to image rotation as well as to linear change of illumination. In addition, experimental results show also that these two descriptors perform well under image affine transformation, blurring, JPEG compression and nonlinear change of illumination.
修改评论