Object Contour Extraction Using Adaptive NURBS VFC Active Contour

Abstract

We proposed a novel active contour model, adaptive NURBS VFC snake, for extracting complex object boundaries. This model enables local control as well as adaptive shape description capabilities for state-of-the-art VFC (vector field convolution) snake to capture complicated contours with high concavity regions. The active contour is expressed as a non-uniform rational B-spline (NURBS) with a novel approach to iteratively estimate the control points' positions and weight values in the snake' evolution process based on a two-steps linear NURBS fitting method. A proposed algorithm automatically defines the fit number of control points based on the Euclidean distance between two consecutive control points and their weight values. Experimental results demonstrate the capability of accurate adaptive shape description for complex object contour with high concavity regions, and advantages in capture range, and initialization insensitivity.