Extraction of shape elements in low-level processing for image understanding systems

Extraction of shape elements from images is one of the most important tasks in scene analysis systems. In the approach presented in this paper the extraction of shape elements is based on the polygonal decomposition of the interior of the object in the scene and on the description of its boundaries (edges). The interior decomposition is done using squared polygons (prime components). The boundaries of the object are determined using a modified Sobel operator. After the decomposition of the interior of the object and detecting its boundaries, a discrete set of points representing the edges and coordinates of the prime components is processed by a nonlinear algorithm. The purpose of this preprocessing is to remove the discontinuities and to produce a discrete skeleton of the object. The final stage of the extraction is to transform the discrete skeleton of the object into its shape description (shape descriptor). This process is carried out by matching the B-spline curves to the whole discreet skeleton of the object. The resulting shape descriptor of the object is composed of a set of extracted shape elements (set of B-spline curves) and an interrelation among them