Hybrid approach for extraction of dominant points from digital curves

In this investigation, we propose a scheme which integrates the two complementary approaches, i.e., the raster approach and the vector approach, to minimize the number of feature points but still preserve the shape of digital curves with high fidelity. The raster form of the digital curve is first encoded in the Freeman chain code space. Then the second order derivative operator combined with a Gaussian filter is applied to detect zero crossings. The turning points on the curve detected as zero crossings in Freeman chain code space are then transformed to the image space. The second stage is then applying dynamic strip algorithm to further eliminate unnecessary points remaining in the first stage. In order to avoid over- elimination of the points, the maximum length of the strip may also be assigned in the scheme. To quantitatively analyze the performance of the proposed scheme, the perpendicular distance of each point on the original curve to the fitted line segment characterized by the dominant points is calculated. The applicability of the proposed scheme in surface modeling from a contour map is also tested.