3-D range images segmentation based on Deriche´s optimum filters

Recursive filtering appears as an efficient tool to implement a robust multiscale edge detection schemes. In this context, and based on Canny´s (1983) design, Deriche (1987) propose a set of optimum filters {S.χ.e^-α.|χ|}_αεℛ, of which the performance´s degree depends on the choice of the parameter α´s value. In the first part of this paper, by theoretical error analysis, we determine the best “range” of α´s values. A matrix representation of the recursive system is also presented. From the first part results and using Deriche´s design, the second part presents a segmentation procedure, using the 3-sign curvature label scheme, where the first, second and mixed derivatives used in the determination of both Gaussian and mean curvatures, are calculated