Contour Segmentation Algorithm of Multi-scale GVF Snake

Snake has two major difficulties: one is the very narrow capture range and the other is the difficulty in moving into boundary concavities. One of its advances, gradient vector flow (GVF) Snake has the advantages of insensitivity to contour initialization and its ability to deform into concave part of the object compared to other deformable contour models. However, the performance of a GVF snake to model any arbitrary shape is heavily dependent upon objects with the highest intensity changes in the edge map, rigidity parameters´ selection and having an uneven spacing problem. To alleviate these problems, a new contour extracting method, GVF Snakes combined with multi-scale Gaussian filter, is proposed in this paper. In this algorithm, in order to increase the capture range of the snake, the image is filtered by a two-dimensional Gaussian kernel with standard deviation sigma. Then, using a gradient vector flow model (GVF Snake) for the external force and increasing or decreasing snake points when it is necessary, sigma is changed in the order of degressive scale before the multi-scale GVF Snakes is used every time to extract accurate contour of target. Meanwhile, a Canny edge algorithm is applied to obtain the initial edge map and the dynamic expanding filter is used to eliminate the noise in the image. The experimental results for synthetic image and real image indicate that the method proposed in this paper is better than GVF Snake algorithm in fitting for deep boundary concavities and simplicity of parameters´ selection