Research On The Segmentation Algorithm For Fiber Cross-sectional Image

Automatic fiber recognition deals with many research fields such as image processing, pattern recognition, computer vision and neural networks. Because the manual fiber recognition is time-consuming and its accuracy is largely depended on operators, the computer aided fiber recognition has received much attention. However, the algorithm for automatic fiber recognition is complicated, and no research result has been applied successfully. This paper is a part of the research sponsored by the Foundation of National Excellent Doctoral Dissertation of China and the Scientific Research Foundation for the Returned Overseas Chinese Scholars, State Education Ministry. The research is also sponsored by the Shanghai Entry-Exit Inspection and Quarantine Bureau of PR China. The research mainly deals with the algorithms for automatic recognition of the nature cellulose fiber and the classification of shaped fiber.Image segmentation is an important pre-processing for the fiber recognition. The bias of the segmentation will largely affect the accuracy of fiber's feature recognition. Fiber overlapping inevitably exists in the fiber cross sectional images. In this paper, many image segmentation algorithms are reviewed, and their limitations for fiber separation is studied. The distance transform based algorithm for fiber separation is proposed in the paper.In the proposed algorithm, the projection is established between the input binary signal set and the elements set for segmentation result; the processing of region filling and boundary tracking is also applied; with the distance transform as the core operator, the classification of the fiber elements set according to the projection is defined, and all the fiber element-sets are counted. In the algorithm, the mask set is calculated for the elements with output (value=1) in the binary image. After that, the distance transform for the input image is executed with the target of the mask set. Region filling takes place from the initial points with large distance transform result. The background set is recognized and it is composed of the elements without output (value=0). The region filling is also processed for the elements without output sequentially. These filling results represent fiber lumens. Finally, a distance transform is computed for the mask set with each lumen set as the target. Each element of mask set is associated to the closest lumen set as the result for the fiber image segmentation.With experimental results, in the proposed algorithm, no bias occurred to the separated fibers compared to the conventional algorithms in literatures that separate objects based upon the contour calculation solely. Satisfactory result is obtained for the application of mass overlapped fiber separations.