The Research And Application Of Overlapped Profiled Fiber Segmentation Based On Watershed Algorithm

Profiled fiber mixing ratio is an important quality index in examining the textile product, and has always been valued high by quality inspection stuff. Traditionally, profiled fiber detecting depends on manual or semi-manual which has many defects such as time-consuming, low precision, poor data stability. The application of computer image processing technology in profiled fiber inspection improves recognition rate and accuracy. However, it is still very difficult to apply computer technology to automatic profiled fiber identification. Since the research is very less, there are still many problems to be resolved.For the purpose of automatic profiled fiber identification, the profiled fiber inspection has five steps which are image capturing, image pre-processing, image segmentation, feature extraction and pattern recognition. This paper studies the demands of Shanghai Entry-Exit Inspection and Quarantine Bureau on textile and fiber test and focuses on the research of automatic profiled fiber identification. This paper pays attention to segmenting the overlapped profiled fiber.With the impacts of profiled fiber slice crafts, micrograph facilities and so on, there are many overlapped profiled fibers among the collected sample images. Profiled fiber image segmentation, especially the overlapped profiled fiber image segmentation is the key step of the whole process in automatic profiled fiber identification.. Therefore overlapped profiled fiber segmentation is the guarantee of feature extraction, profiled fiber classification and profiled fiber counts.Firstly, this paper analyzes and sums up the principle of algorithms in overlapped image segmentation, and discusses the advantages and disadvantages of the each algorithm. Considering the profiled fiber's characteristic, the limitations of common algorithms are pointed out, and then a new algorithm based on Watershed algorithm is proposed.Watershed algorithm is extremely sensitive to noise, which would make the overlapped profiled fiber image over-segmented. To solve this problem, this paper uses Triangle Area Representation (TAR) to mark the feature concave points. Then it deletes the watershed lines which didn't go through feature points with the adaptive slide window strategy. The correct watershed lines would be retained as well as redundant ones are eliminated.The algorithm analysis and experimental results show that the proposed algorithm could overcome the over-segmented of traditional watershed algorithm and could accurately segment the overlapped profiled fiber. Satisfactory result provides reliable data for follow-up processing of automatic inspection and identification profiled fiber.