Edge detection at subpixel accuracy via ideal interpolation.

摘要

This dissertation presents two novel techniques for edge detection at subpixel accuracy. In the first method, a windowed version of the ideal interpolation formula is applied to an image, from which the first- and second-order derivatives are used to localize edges in order of strength. The second method applies the derivative of the windowed ideal interpolation formula to the image, from which direct edge localization is performed. Once an edge is located, the entire edge may be linked, eliminating the need for thinning. The proposed edge detector converts roof edges into step edges, eliminating the handling of multiple edge types, and by interpolating, the effect of the integer grid problem associated with the sampling process is reduced.