The Research And Application Of The Edge Detecton Algorithm Based On Grey System Theory

Automatic fiber identification is involved in many scientific research fields, such as digital image processing, pattern recognition, computer vision and grey system theory and so on. With the improvement of people’s living quality and the scientific progress, more and more kinds of fibers appeared in the international market. And the composition of fiber fabric affects the performance and price, so it becomes more and more important to recognize the ingredient of the fiber fabrics. As the traditional artificial or half artificial fiber identification methods existing many drawbacks and the rapid development of computer digital image processing makes the computer aided fiber identification system possible. However, the automatic fiber identification is a complicated problem, and there are still many problems to be solved.During the fiber embedding, slicing and acquisition process, the equipment constrains, technology restriction and artificial factors make the obtained fiber images have many defects, such as uneven illumination, edge blur, serious noise pollution, the ambiguous distinction of target and background, adhesion, etc., which brings a big problem for the fiber separation, feature extraction and recognition. So the profiled extraction of the fiber images is very important. And it is necessary to research an effective edge detection method with robust to extract the edge of the low-qualified fiber images.After doing research on fiber image with low quality and the grey prediction model, the grey correlation degree, directed graph and existing edge detection algorithm, this article proposed a new edge detection algorithm to obtain complete and continuous edge. It mainly includes:Due to the collected fiber images exist serious noise pollution, which got by embedding, slicing and enlargement, and we cannot get good result if directly apply edge detection algorithm. So this paper uses Gaussian Filter to suppress the unwanted noise.According to the characteristic of the collected fiber images, such as unobvious distinguish between the target and background, adhesion and edge blur, this article obtains the fiber outline firstly, and then the edge detection algorithm is applied to the adhesive fibers. First of all, the outer contour extraction algorithm is proposed based on the idea of double scale image edge detection. The grey correlation degree is applied to extract Region of Interest (ROI) based on its application in image processing, the grey prediction model is used to get strong edge because it can locate accurately, and the Niblack algorithm is designed for gaining weak edge on account of it can receive rich edge. This method can get the complete and continuous edge information, but cannot eliminate the false edge inside the fiber. Thus, filling algorithm and contour tracking algorithm are used to get the fibers’ outer contour, which can obtain the single or adhesive fiber’s outer contour.Then, according to the advantages and disadvantages of the directed graph this paper combines the above algorithm with the original image, and uses the mean of the image to fill the background area, so as to achieve enhanced fiber image. And then apply the directed graph algorithm to edge detection. Thus we not only can get the fuzzy edge, but also ensure the edges’ continuity and integrity.Finally, obtain the edge of single pixel value by refining function, and then use the contour tracking algorithm to eliminate the burr, and the final edge is appeared.Proven by the experimental results, the proposed edge detection algorithm can overcome the defects of the conventional edge detection algorithm, such fracture as edge, false edge, etc. This algorithm not only can extract the edge of the fiber image accurately, but also can get the edge of the adhesive fibers effectively, which provide a guarantee of the subsequent separation and identification.