Generalized Hough Transform to Be Extended as an Affine-Invariant Detector of Arbitrary Shapes

摘要

The problem of detecting an arbitrary curved shape from an image in an affine-invariant form is an important basic research topic in image recognition. Although several conventional techniques for achieving this kind of detection goal have been proposed, none of them are able to deal with the following kinds of situations very well: (1) part of the contour of the shape to be detected is hidden due to overlap with another object and (2) breaks occur in the contour due to noise. Therefore, in this paper, the authors propose a new shape detection method that can even deal with these kinds of situations. First, to stably extract tangent information from a discontinuous contour shape that includes arbitrary curves, they propose a new tangent information extraction method based on arc fitting using a Hough transform. They then focus on the property that the "parallelism of straight lines is preserved under an affine transformation" and actively use tangent information at various contour point positions of the shape to extend Ballard's generalized Hough transform [1] as an affine-invariant shape detection method. The authors performed evaluation experiments to verify the effectiveness of the proposed technique and obtained good detection results.