Abstract: A novel active contour model was developed for the problem that the Gaussian distribution fitting model is sensitive to a starting contour. According to the characteristics of the local intensity clustering， the bias field and a piecewise constant function were integrated to approximate the local image intensities， which made the objects be segmented with the starting contour being anywhere in the image. An efficient numerical schema was used for the implementation of the proposed model in order to converge rapidly and avoid re-initialization. Experimental results on a series of real and synthetic images demonstrate that the proposed model is robust to the starting contour and the images with intensity inhomogeneities can be effectively segmented.

Key words: image segmentation, bias field, piecewise constant, level set, active contour