Fiber Image Segmentation Based On Snake Model

Fiber image segmentation is an indispensable step in the automatic fiber recognition system, and it is the base of the processes of feature exaction, classification recognition. So to segment a complete, continuous and single pixel fiber is the critical task. With the limits of fiber embedding and sectioning technology and illuminated with point light source techniques in making fiber cross-sectional samples, the fiber images sometimes have problems such as distorting and uneven illumination. This situation brings fiber image segmentation many difficulties.Based on the research of classic image segmentation methods, for the limits of traditional segmentation algorithms such as false edge, double edge or discontinuous contour, Snake model is used in fiber image segmentation.The paper illuminates the foundation of Snake model, because Snake model cannot solve boundary concavities better and the original contour must be placed close to the real boundary of object, we introduce GVF (Gradient Vector Flow) Snake model. It can solve these problems better than Snake model. But there are also many difficulties in the fiber image segmentation, such as time efficiency or effective original contour. For these difficulties, the GVF Snake model is applied in the segmentation of fiber image, we analyzes and researches the pre-segmentation processing first, and then a new fiber image segmentation algorithm based on clustering segmentation and GVF Snake model is proposed. The clustering segmentation is used to obtain the original coarse contour of fiber; and next GVF Snake algorithm is applied to calculate the accurate fiber contour. At last due to the noise of fiber micrographic image, some fiber contours have burrs, which can be removed by contour tracing method.The experimental result shows that this algorithm is effectively and accurately, which can not only solve the problems such as false edge, double edge or discontinuous contour of the traditional segmentation algorithms, extract the complete and continuous fiber contour, but also depress the noise of fiber image.