Novel Numerical Methods for Efficient and Reliable Segmentation

Abstract

This article is concerned with efficiency and reliability issues related to level set-based segmentation methods. Geometric active contour methods show some desirable characteristics: flexibility in the topological changes of contours, capability of detecting interior boundaries, and a low sensitivity to noise. However, they tend to detect undesired boundaries when applied to general images. In order to overcome the drawback, we introduce the method of background subtraction (MBS), which transforms a general image to an essentially binary image and therefore conventional segmentation methods can detect desired edges more effectively. An effective initialization technique for the level set function and a hybridization of information from both the intensity and statistical properties (distributions) are also introduced to improve efficiency and reliability of level set-based segmentation methods. The resulting algorithm has proved to locate the desired edges in 2-4 iterations, for various images.