Local Region Active Contours Based on Gauss Smoothing
Résumé
Local region-based active contours using Gauss smooth function to approximate images are very sensitive to initial localizations and to image shapes. In this paper, we propose a coarse segmentation stage which maps the local-region data inside and outside the closed curves to two constants respectively. The constants can be computed by minimizing one-dimensional energy function and they are used to approximate the image to avoid falling into the local minimum. Experiments demonstrate that our proposed method makes a significant improvement both in synthetic and natural images.