Pixel-jump fast image matching algorithm based on LTS-HD

By analyzing the characteristics of the HD measure, a real-time two-level scene matching algorithm is proposed. Compared with traditional image multi-scale feature decomposition matching methods, the proposed method is performed directly on the original resolution image and shortens the matching time by reducing both the number of match points and the computation complexity of similarity measure. The pixel-jump method decreases the participated match points greatly, and the computation of LTS-HD at each point is performed between sets consisted of only feature points, which efficiently simplify the computation of similarity measure. To ensure the match precision, a coarse-to-fine two-level matching strategy is adopted. In the first level, a coarse match point is obtained using pixel-jump searching through the reference image. In the second level, a point-by-point local searching is performed to get the accurate match point within the δ-neighborhood around the coarse point. Simulation results show that the proposed matching method takes less time than both the point-by-point searching and the genetic algorithm, and that the match point is correct even though the actual image is occluded severely.