Abstract:

In order to identify the different components which are adhered or crossed each other in a hybrid fiber image, a fiber segmentation algorithm based on concave points matching in the overlap area is proposed. The detection of contour corners is firstly obtained by using the classical K-cosine curvature algorithm, the vector triangle area method is also used to find contour concave points from the corner set. For a given concave point, the precursor point (or subsequent point) and the target point are selected to form a triangle. The triangle and other significant features are used to judge whether the concave point and target point are matched. At last, through the connection of the matching concave points, the segmentation of overlapped fiber image can be realized. Compared with the existing algorithms of segmentation, the new method employs a mechanism combined with distance between concave points and triangular construction principle to achieve the purpose of matching. The experimental results show that the algorithm can deal with the situation of adhesive fiber and cross fiber, with high segmentation accuracy over 80% in complex scenes.

Key words: fiber image segmentation, overlapped fiber, corner, vector triangle area, concave point matching