Automatic Grade Evaluation Research For Fabric Pilling Based On Image Analysis

During the wearing and washing, pills caused by friction will be appeared on the fabric surface. The pills on the fabric surface not only damage the fabric appearance, but also deteriorate the touching of fabric. This will damage the wearing properties of fabric. During the inspection of fabric or clothing, the resistance of fabric pilling is adopted as one of the important measurement properties for evaluating fabric quality. The existing grade evaluation method for pilling fabric based on standard photos of pilling fabric, is relied on the human vision. The inspection process has the flaws of subjective and the grade evaluation result is partly random. To make the grade evaluation of fabric pilling to be objective and reliable, a novel method for automatic evaluate the grade of fabric pilling is proposed in this paper. The research devotes to development an automatic grade evaluation system for fabric pilling.The research of this paper includes major contents as follows:automatic recognition of fabric structure parameter for grade evaluation of fabric pilling; the fabric pill information extraction based on Fourier transform; automatic analysis of fabric pills with Gabor transform in time-domain; fabric pills detection with Gabor transform in frequency-domain; characteristic parameters extraction for evaluating fabric pilling grade; the research and discussion of fabric pilling evaluation method; the development of automatic fabric pilling evaluation system.In Chapter 1 the background and significance of the topic selection of the thesis is introduced briefly. First, The simulation methods of fabric pilling caused by friction during wearing and washing in the lab are introduced and the existing research content and analysis method are discussed. Then, the research of the grade evaluation for fabric pilling based on digital image analysis home and abroad is summarized. The shortcomings of the automatic grade evaluation of fabric pilling based on image processing are analyzed in conclusion. The research objects, content and major innovations of this paper are given at the end of this chapter.In Chapter 2 the automatic recognition process of fabric structure texture parameters used for evaluating the fabric pilling are discussed. First, the definitions of relative fabric structure parameters are introduced. Based on the discussion of the relation between the parameters and composition of pill.ng fabric image, the fabric density and.he number of fabric weave repeat are chosen as the parameters for determining the structure composition of fabric image. The inspection of fabric density with gray-projection method is discussed and the procedure of the inspection is given. To smooth up the affection of noise information, the fabric image is transformed into frequency-domain with Fourier transform. The amplitude spectrum of fabric image is then obtained and analyzed. The filtering method for separating the warp and weft yarns in the frequency-domain is proposed by selecting the region representing the period information of warp and weft in the fabric image. In the sub-image including separation yarns, gray-projection method is used to locate the warp and weft yarn and the fabric density can be measured. To detect the number of fabric weave repeat, the yarns in the fabric image are segmented based on the fabric density detection result. The projection information along the longitudinal direction is detected to characterize the separating yarn. The number of fabric weave repeat can be obtained by calculating the relation of the gray information of different yarns.In Chapter 3 the extraction method of fabric pills with Fourier transform method is given. The gray projection character in fabric image is introduced first and the gray intensity in the yarn obeys the sine curve. Based on this, Fourier transform which is based on sine function is proposed to analyze the fabric image. It is then selected to remove the periodic signals in the fabric image. The image acquisition method with a flat scanner of pilling fabric is given and the standard pilling image of wool woven fabric in GB/T4802.3 is chosen as the analysis object. The periodic signals in the fabric image are chosen with the help of amplitude spectrum obtained by Fourier transform and they are eliminated by threshold segmentation in the frequency-domain. The low frequency signals are retained to enhance the fabric pill information. The fabric image with enough difference between the background and the pill are got with the Fourier analysis method. The enhanced local Otsu threshold method is last adopted to segment the fabric pills automatically. The segmentation for standard pilling fabric image of five grades proves that the method pills based Fourier transform can be used for extracting the fabric pills.In Chapter 4 the method of extracting pills in pilling fabric image based on Gabor transform in time-domain is discussed. The definition of Gabor transform is introduced first and the 2D Gabor filtering method in time-domain is explained. The parameters of Gabor filter, including the direction parameter, proportion parameter, frequency parameter, scale parameter and the window size of Gabor filter are then described with typical Gabor filters and the selection way of these parameters is given then. The size of the Gabor filter is determined by the region of the fabric structure in the fabric image. Based on the frequency parameter and direction parameter, a filter bank including 16 Gabor filters is constructed. To obtain satisfied Gabor filtering result, the selecting rule of optimal Gabor filters is given by calculating the mean and standard deviation of filtering result. The fusion of Gabor filtering result is carried out by fusion of the Gabor filtering results of different directions in the same frequency. The direction of fabric pills in the filtering result can be then eliminated. Based on the Gabor filtering and fusion result image, dual threshold values constructed by the mean and standard deviation of the result image are used to segment the fabric image. The pills can be located with threshold segmentation. The fabric pilling images of different grades are filtered by the proposed Gabor filters. The filtering results proved that the Gabor filters used can eliminate the fabric structure textures and enhance the fabric pill information. In the filtering image, the fabric pills can be segmented with simple threshold method.In Chapter 5 the method of extracting pills in pilling fabric image based on Gabor transform in frequency-domain is discussed. The definition of Gabor transform in frequency-domain is first introduced in the chapter and the filtering way for image of Gabor transform in frequency-domain is explained. The realization method of Gabor transform in frequency-domain is then given and the parameters, including the scale parameter, the proportion parameter and the coordinate parameter, are discussed with the affection to the character of Gabor filters. The basis selection rules of these parameters are given then. To get satisfied Gabor filtering results, the different Gabor filters are adopted to filter the given pilling fabric image. Based on the filtering results, the coordinate parameter is selected and the results in different directions are fused to remove the direction affection of fabric pills. The selection rule of Gabor filter used for extracting the fabric pills is then given by the peak filtering method and the fabric density detection results. In the filtering result, local threshold method is chosen to locate and segment the fabric pills. The fabric pilling images of different grades are filtered by the proposed Gabor filters in frequency domain. The filtering results proved that the Gabor filters used in this chapter can eliminate the fabric structure textures and enhance the fabric pill information. Compared to the Gabor filtering method in time-domain, the Gabor filters in frequency-domain selects the filtering parameters based on the fabric weave structure parameters. The better filtering result can be obtained and the pill information in the fabric image is extracted accurately.In Chapter 6 the characteristic parameters of fabric pilling and the method used for evaluating the fabric pilling grade are discussed. The pilling grade evaluation system constructed is also introduced. Based on the pill segmentation results, the characteristic parameters can be calculated with statistical method. Five parameters, including the area of fabric pills, the number of fabric pills and the density of fabric pill, are adopted to evaluate the fabric pilling degree. The relation between the parameters and the fabric pilling grades is analyzed and the total area of fabric pills is selected as the evaluation standard for fabric pilling. To eliminate the magnify affection of fabric image, the scale of the total area of fabric pills to the area of fabric image is chosen as the evaluation index. Based on the scale of total area of fabric pills, two methods are used to assess the fabric pilling grade. In the first method, the cutoff values between different grades are calculated with interpolation. The fabric pilling grade can be then determined based on the scale of total area of fabric pills to the area of fabric image. The second method is to assure the function relation between the scale of total area of fabric pills and the fabric pilling grades. The scale of total area of fabric pills can be used as a parameter to calculate the fabric pilling grade directly. In the latter part of this chapter, the construction of grade evaluation system for fabric pilling is explained. The hardware and software of the system is introduced. The industrial camera SCA 1600 produced by Basler corporation in German with IEEE 1394 port is selected as the hardware of image acquisition. Visual Basic 6.0 and the IMAQ ActiveX of American National Instruments are chosen as the software to construct the fabric pilling automatic evaluation system. The realization for the functions of the fabric pilling evaluation system is then briefly introduced and the reference interface of the system is provided at last.In Chapter 7 a summary is made to describe the main contributions and the problems of the present work. The advice of the future work of the automatic evaluation of fabric pilling based on image processing is given at last.