Contour-based shape representation using principal curves

Extraction and representation of contours are challenging problems and are crucial for many image processing applications. In this study, given a membership function that returns the score of a point belonging to a contour, we propose a method for contour representation based on the principal curve (PC) of this function. The proposed method provides a piecewise linear representation of the contour with fewer points while preserving shape. Varied experiments are conducted, including lung boundary representation in CT images and shape representation in handwritten images. The results show that the technique provides accurate shape representation. ► We extract the contour given a membership function describing the shape. ► Contour points are down-sampled creating a piecewise linear representation. ► Proposed representation is used for images in different domains. ► The experiments show that the technique is insensitive to image noise. ► The technique preserves shape to a user-defined accuracy.