Research On The Segmentation Algorithm For Overlapped Fibers

The characteristics and mixture ratio of fibers are decisive to the quality of apparel products. The kinds of fibers are mainly cotton, wool, silk, and ramie. The classification of textile fiber is important in import and export inspection. It is carried out manually at present. Manual inspection is time-consuming and costs much labor power. Moreover, the inspection result is unstable. Therefore, much research has been devoted to computer-aided fiber inspection. Fiber character extraction is one of the important preprocessing.In fiber samples, fibers are inevitably overlapped and intersected. Fiber character extraction is implemented upon well-separated fibers; hence, the algorithms for fiber separation and single fiber extraction are proposed in the thesis.Conventional fiber separation algorithms are implemented mainly with 2D images. Branches are grouped by the comparison of branch slope. Therefore, it is difficult to distinguish the fiber overlapping or bifurcation. Based upon the variation of definition with multi-focus plane for the same scene, the layer number is determined for each pixel in the scene as Z direction parameter. Pixels of same fiber have the similar Z direction parameters, while pixels of different fibers especially in the crossing part have different Z direction parameters. Fiber separation is carried out with 3D information (X, Y, Z).In the separation algorithm proposed in the thesis, images for the same scene are taken with sweeping focus plane in Z direction. All fibers in the scene should been covered in the sweeping range. The value of definition is calculated as the variance of fiber pixel in its 8 neighborhood. The definition value is calculated for each pixel in corresponding location in each image under different focus; the layer index with maximal definition value is determined. And then the background separation, impurity removal, and denoising is implemented. With the layer index matrix, the layer number of each fiber segment is obtained. The fiber skeleton is calculated through thinning table; less important skeleton information is pruned; effective fiber skeleton part is classified; public skeleton segments and construct matching table are obtained. Based on the definition value of each pixel, the definition value is increased at fiber outline. With the relationship of fiber outline and fiber skeleton, the fiber skeleton line is set in match table. Fiber skeleton line is matched with the minimum differences of layer number. With above processing, fibers are separated. This algorithm is demonstrated to be effective for fiber overlapping and bifurcation.