Affine-invariant detection of arbitrary 2D objects/shapes from images is one of the most important and fundamental issues in image recognition. In this paper, we propose a method to realize such a detection task, which uses affine-invariant ratio ρ and least-squares method (LSM). We focus on one of the geometric properties of 2D affine transformation whereby the length ratio of line segments on the same line is preserved under any 2D affine transformation, and then we adopt the affine-invariant ratio to realize fast access to the reference table. We did several experiments to verify the effectiveness of the proposed method, and as a result, we confirmed that it is available for broken shapes whose boundaries are continuously occluded by other overlapping shapes or noises, and is also sufficiently practical.
抄録全体を表示