Segmentation of cross-sectional images using fuzzy logic

The authors present a new segmentation method for range images consisting of a set of planar cross-sectional contours. The approach is novel in that it uses fuzzy criteria for grouping primitives and identifying homogeneous regions. The method was applied to images provided by a structured light sensor. In this case, the image sequence corresponds to the scene profiles obtained with the successive positions of a rotating plane of laser light. It is assumed that the object surfaces can be modeled by a set of quadratic patches. The primitives used for region segmentation result from the approximation of the light profile by second-order curves. An efficient tracking of these noisy curves is achieved by using a fuzzy decision-making algorithm. Region growing is then performed by matching 2-D curves from the image sequence. Results are presented for real scenes consisting of multiple objects of arbitrary shapes. They show that an efficient surface segmentation may be obtained with few-constraint environments including planar or curved shapes