The Research Of Preprocessing Technology For Profiled Fiber Images

With the improvement of living standards and scientific technology, the fiber performance of conventional section can not meet people's needs any more, which leads to the appearance of more and more profiled fibers. However, how to classify different profiled fibers is a important research topic of current textile identification field.As traditional profiled fibers detection methods have many drawbacks, there has been great progress in researching automatic identification of profiled fibers with the development of computer image processing technology.However, during the process of fibers embedding, slice and micro collection, the equipment environment constrains, technology limitation and human factors make the collected sample images have many shortcomings, such as poor illumination, low contrast, the ambiguous distinction of target and background, a lot of vague contours, etc. These defects bring huge difficulties to the image separation, feature extraction and recognition during post processing. Therefore, the preprocessing of profiled fiber images is essential, so the study is to find a method to correct non-uniform illumination, overcome low contrast and reinforce the fuzzy outline of the fiber.Firstly, as the production processing limitation of profiled fiber biopsy sample, the non-uniform illumination impacts the fiber contour and the background contrast, this paper proposes a new algorithm combining average filter with high-pass filter. The experiments prove that this algorithm can not only balance the light inequality better, but also improve the profiled section contour and background contrast. Besides, the algorithm's structure is simple, without multiple iterations.Secondly,for noise and fuzzy contours phenomenon, the paper proposes an image enhancement method that combines Canny operator, contour pattern, and a nonlinear S function with adaptive threshold, which could not only strengthen the target area but also suppress the background noise. At the same time, this algorithm limits the enhancement region, which effectively improves the efficiency. The Otsu binarization process of the enhanced images shows that the algorithm can not only enhance the contours of original sections, but also can enhance those fuzzy or even break contours, which make the obtained contours clear and complete, and have a very strong versatility. Lastly, during the process of the de-noising and enhancing algorithm, some noises are enhanced to variety degrees. Some even have glitches and fractures on the contour line of profiled fiber image, which inevitably affect the contour separation, feature extraction and completed sample number of profiled fiber. For these defects, a contour tracking algorithm to eliminate the edge of the branch is proposed, and a repair model according to the classification of gap distance in broken contour lines. Through the above methods of branches and glitches eliminating and fracture contour lines repairing, a complete and smooth contour is obtained, which finally provide a strong guarantee for later overlapped fiber isolation and features extraction.