Subdivision based snakes for contour detection

TL;DR

This paper introduces a novel subdivision curve-based snake method for efficient and robust object contour detection, leveraging hierarchical subdivision properties and a new region energy for contrast maximization.

Contribution

It presents a new subdivision curve snake approach with a hierarchical optimization process and a novel contrast-based region energy for improved contour detection.

Findings

Method is fast and robust in experiments.

Effective with synthetic and real images.

Utilizes hierarchical subdivision for efficient computation.

Abstract

In this paper we propose a method for computing the contour of an object in an image using a snake represented as a subdivision curve. The evolution of the snake is driven by its control points which are computed minimizing an energy that pushes the snake towards the boundary of the interest region. Our method profits from the hierarchical nature of subdivision curves, since the unknowns of the optimization process are the few control points of the subdivision curve in the coarse representation and, at the same time, good approximations of the energies and their derivatives are obtained from the fine representation. We introduce a new region energy that guides the snake maximizing the contrast between the average intensity of the image within the snake and over the complement of the snake in a bounding box that does not change during the optimization. To illustrate the performance of our method we discuss the snakes associated with two classical subdivision schemes: the four point scheme and the cubic B-spline. Our experiments using synthetic and real images confirm that the proposed method is fast and robust.