Efficient contour representation and estimation using image segmentation and second-order B-spline descriptors

A novel algorithm is presented for segmentation and approximation of object boundaries using edge detection and periodic second-order B-splines. A rectangular region of interest containing the closed object boundary is defined. Segmentation is done and target contour is identified followed by smoothing and the extraction of geometrical properties. Iteration is minimal and the error criterion is defined as a 1-pixel tolerance band around the contour. Unlike other active contour algorithms, the order of parameterization is unsupervised and the only a priori condition is the rectangular region of interest. The computational efficiency of this algorithm surpasses that of other active contours. The experiments reported in the paper, performed on MR (Magnetic Resonance) images, illustrate the adequacy and good performance of this approach.