Navigation image matching based on two-level keypoints

A fundamental problem in scene matching is the precise determination of correspondences between pairs of images, especially used in the cruise missile. Many methods have been proposed and work very well for image data in personal computers. However, with the speed of cruise missile becoming faster and faster, the time consumption of matching methods are too large and cannot be equal to the matching need in the scene matching of cruise missile. This paper proposes a novel matching framework based on a new two-level keypoints structure to find points and local image regions correspondences between pairs of images. Harris corner detector is used to extract the two-level keypoints with different detecting radius (low-level keypoints´ detecting radius is small and high-level keypoints´ detecting radius is large). Then two new feature descriptors are proposed according to the two-level keypionts. Low-level feature descriptors are formed by sampling the pixel values around the low-level keypoints. High-level keypoint can represent a local image region, whose center point is the high-level keypoint. The local region can be divided into several ring parts by different radiuses. Every high-level feature descriptor is formed by counts the numbers of low-level keypoints in every ring parts, which belongs to the local image region. We additionally optimize the matching regions by a vote approach.