Abstract:

Active contour is an important method to segment images. Its segmentation strategy is usually edge-based or region-based and based on the contour energy minimization. But the classical active contour is difficult to deal with the region which has weak edges or is in the complicated background. In this paper, we implement an improved statistical region-based active contour and propose a strategy to extract region of interest by polygon. Within the family of exponential probability density function, statistical region-based active contour can run by fast algorithm. We give details of the realization of the algorithm and simplify the contour by the minimum description length. Experiment results illustrate the efficiency of our proposed method in contour extraction.

Key words: statistical active contour, planar features, region extraction, random search, minimum description length