Homography based quasi-dense match propagation for fisheye images

Most of the quasi-dense matching algorithms are designed for general scene structures. In reality, there are many man-made objects in the scene, such as for example buildings and furniture which include many planar regions. For this case, plane induced homography is a useful tool for matching. In this paper, we present a novel quasi-dense matching method for fisheye images which is based on plane homography. First, sparse planar seed regions are detected and matched by using affine invariant detector and matcher. Then region growing algorithm is employed to propagate seed region to its neighboring pixels by the best-first strategy which greatly improves the robustness and stability with regard to the sparse seed regions. We also give a new criterion for evaluating the goodness of seed regions. Finally, some wrongly propagated points are checked and re-allocated to their correct positions. Experiment results showed its feasibility.