The Evaluation Of Pilling Grade Of Fabrics Using Computer Vision Based On Tangential Projected Images

The objective evaluation of fabric pilling grade based on tangential light projected image using computer vision is studied in this thesis. The main contents covered by the thesis are the design and installation of the device for capturing light projected images, the extraction of the tangential projected profile of pilled fabric surface and the development of 3-D image of pilled fabric from the integration of sequential tangential projected profile, the algorithm and results of the image preprocessing and pill segmentation, the extraction of pill's feature index, the assessment of pilling grade by neural network, and etc. The following is a brief introduction of the content for each chapter.In foreword the background of topic selection of this thesis is described briefly.In chapter 1 a brief account of the recent progress in the world of studies on the evaluation of pilling grade using computer vision is presented. In addition to the general description of the method of pilling testing and the subjective evaluation of pilling grade, the new results and deficiency of the objective evaluation of pilling grade using computer vision, which is explained by both the gray image analysis and range image analysis, are given out emphatically.In chapter 2 the development of the device for capturing the projected images and the design of shaped bar which is the key part of the device, are discussed. And it is also analyzed in detail that on the images of pilled fabrics how the ratio of the edge curvature radius of shaped bar to pills radius affects the distortion of pills either extension or shorten and the conglutination of pills, ie, the projected image of more than one pills superimposed along the lines of sample moving forward. The cross-sectional figure of shaped bar is considered properly according to the requirements of both structure and rigidity.In chapter 3 the development of pilling images of pilled fabric samples based on tangential light projected images is elaborated. The main contents include as follows: the investigation of background light intensity for capturing the projected images with proper contrast, the detection of the tangential projected profile of pilled fabric surface using edge detection of LOG operator and liner interpolation, the acquiring of the 3-D image of pilled fabric by integrating sequential projected profiles, the 2-D gray display of the 3-D image and etc. The method of acquiring the pilling image of fabric samples described in this chapter can eliminate the interference with pilling information from fabric color and pattern, and therefore can be applied to not only solid colored fabric, but also colored woven fabric, printed fabric and other fancy fabric.In chapter 4 the image preprocessing and pill segmentation are discussed. The background unevenness and the pill's gap on pilling images of pilled fabric are analyzed and their removals using operation of mathematical morphology and other method are introduced in detail. Four different kinds of threshold are used to segment the pills on pilling images. Besides showing all the results, the emphasis is put on the analysis of the reasonableness and feasibility of pill segmentation by threshold based on Gaussian model. Finally, the results of pill segmentation with reference to the pilling images of solid color fabric and especially colored and printed fabric by the threshold based on Gaussian model are provided.In chapter 5 the extraction of pill's feature indexes, such as roughness, number of pills, total area of pills, fractal dimension of pills, is introduced. The correlation between each feature index and pilling grade and the separability of each index are analyzed particularly, which provides the foundation of assessing pilling grade.In chapter 6 the training and testing process of Bp network and SOFM network using 100 fabric samples, and the assessed results of pilling grade by neural network are presented. The correlation between pilling grade assessed by Bp network and that by skilled person is 0.97, and...