Improved Minimum Distance Discrimination Method Used in Image Analysis of Fabric Wear Resistance

Transcription

1 Applied Mechanics and Materials Online: ISSN: , Vols , pp doi: / Trans Tech Publications, Switzerland Improved Minimum Distance Discrimination Method Used in Image Analysis of Fabric Wear Resistance Jianxia Su 1, a, Runping Han 1, b,limin Shi 2, c and Xinghe Qian 1, d 1 Information Engineering School of Beijing Institute of Fashion Technology, Beijing, China 2 Clothing Art and Engineering school of Beijing Institute of Fashion Technology, Beijing, China a ssybj@sohu.com, b gxyhrp@bift.edu.cn, c cnbjslm@126.com, d @qq.com Keywords: image process, level, wear resistance, MATLAB Abstract. With the increasing attention of textile, the wear resistance of the fabric is more and more important today. The wear resistance is judged by the sample and standard sample according to sensory contrast complete, but this approach has repeated poor consistency. The subjective factors of testers influence the result heavily. So the powerful image processing tools of MATLAB was used in the analysis of fabrics image processing. By image read, display and interception, image preprocessing, and feature extraction value, the eigenvalues of total area of the hair bulb, maximum area of the hair bulb, and hair bulb number were obtained. After the improved minimum distance discrimination method was used to determine the level of wear resistance by computer. Tests show that fabrics wear resistance level can be judged quickly, accurately and automatically. Introduction Clothing fabrics suffer a variety of pressure and the frictional force of the outside world in the process of wearing and using. In addition to a limited number of natural fiber products, most fiber products will appear pilling phenomenon. Wool fiber apparel fabrics easy to tick a single head or ring, the surface of the fiber bring out the now fabric surface, known as "fluff" after repeated friction, hair fibers intertwined to form spherical, that is, said "pilling". There are a lot of determination of method and apparatus of the fabric pilling. Design principles are generally of clothing true when worn pilling formation process simulation. Today, the fabric pilling evaluation method includes subjective assessment and objective assessment. Subjective assessment is completed through the sample and standard sample according to the sensory contrast, domestic use of this method, the methods are generally gaze rating evaluator in the rating box fabric, to be assessed compared with the standard sample according to a five-tier system, fabric grade series the smaller, the more serious degree of fabric pilling, contrary pilling resistance better. However, this method has repeated consistency difference by tests subjective factors and the external environment affect too much, resulting in the lack of objectivity of the evaluation results. The objectivity of the assessment by digital image technology in internal and abroad in recent years some researchers try using digital image processing techniques to objectively assess fabric abrasion resistance, but the study remained in the laboratory research stage. This paper is based on MATLAB digital image processing technology, in-depth analysis and research on fabric abrasion resistance rating mainly involved, the pilling image processing and appearance change in image processing. Image Prosessing In this paper grayscale image processing and analysis methods for processing is used for garment fabric pilling phenomenon, divided into three parts of interception of images, image pre-processing and feature extraction value. Image acquisition According to the processing and analysis, the standard sample photo of plain fabric is used to be example in the research. The test instrument is used in the scanner and the computer. The five level standard sample photos image are scanned by scanner, glossy fabric photo in software is shown in Fig.1. All rights reserved. No part of contents of this paper may be reproduced or transmitted in any form or by any means without the written permission of Trans Tech Publications, (ID: , Pennsylvania State University, University Park, USA-19/02/16,14:23:19)

2 422 Information Technology Applications in Industry Fig.1 The standard sample image in software In actual processing, the standard sample photoes of five levels can t be used directly in the image processing, thus requiring interception of the desired image. Fabric center area to better reflect changes in the relationship of fabric pilling level, so this article interception try to intercept the image of the standard sample according to the center of the image. This paper image interception is a = imcrop statement function imcrop can return a cropped area of the image, so that by moving the mouse, you can adjust the size of the selected area, select the image area. That is, the input image can be manually interception part. Intercepted image is shown in Fig.2. Fig. 2 The image after interception Grayhound Image texture filtering before the need for a unified image storage format to be converted to use some image processing functions, the forthcoming RGB function maps were converted to grayscale image, uniform color to gray-scale, The role is to modify images from the RGB image to grayscale, and return the same type of image with the original image storage.grayscale image is shown in Fig.3. Fig.3 The image after gray Images of wavelet denoising Standard sample according to the quantity of the hair bulb is measured from the ball grade scale, but the presence of noise seriously interfere with the identification of the hair bulb, so that the level of assessment impact, it is necessary to effectively filter out the picture noise. According to a different approach to the wavelet coefficients denoising methods are usually divided into based on wavelet transform modulus maxima de-noising. The wavelet denoising image is shown in Fig.4 below: adjacent scale wavelet coefficients correlation denoising threshold denoising based on wavelet transform. These three methods, wavelet threshold denoising is an easy to implement and the effect is better denoising method most widely used.

3 Applied Mechanics and Materials Vols Fig.4 The image after wavelet denoising The image of black and white Fig.5 shows the result after image blacking and whiting. Image binarization Binarized image is shown in Fig.6. Fig.5 The black-and-white image Fig.6 The image of binarization Negated image Negated operation makes the hair bulb pilling part to white in order to facilitate the calculation of the white spot area. Negated image is shown in Fig.7 below: Fig.7 The negated image Median filter The median filter is a nonlinear signal processing method, corresponding to the median filter is a nonlinear filter. Linear filter reduces the noise smoothly, but also blurs the image edges and details. Nonlinear filter smoother is an improvement, i.e., not all the pixels with the average of its field instead, but take a threshold value, when the difference between the pixel gray values with domain

4 424 Information Technology Applications in Industry average mean Replaced; at the threshold value only when the difference is not greater than the threshold value between the average pixel gray value and its fields whichever own gray value.fig.8 shows the median filtere image. Fig.8 The image after median filtering Pilling characteristic parameters extraction Image feature extraction The image feature extraction refers to extraction of the content information contained in the image, such as targets and the extraction of the texture, and then through the matching and other means to identify, and thereby determine the type of target contained in the image number of valid information. Method of calculating the number of hair bulb Binary image search connected region of pixels is a connected region, and a hair ball is pixels in a connected region. Communicating area search for dividing to the image sequence of the first column encountered pixel of the first nonzero numeral 1, which means that all the pixels within this region by reference numeral 1, pixel number as M1, in accordance with Such Cui, until the last a hair ball connected region, designated N, pixel number is Mn, then the hair ball number, namely N. Calculation of the hair bulb area Area refers to a number of pixels in the binary image. MATLAB toolbox with bwarea function, bwarea function not simply calculate the number of pixels 1, but add different weights for each pixel, the weighted sum of the area of the image. For example, the diagonal of the 10 pixels to be longer than the horizontal lines of 10 pixels, as in bwarea area of the horizontal line 10, and the two diagonals of the area of The percentage calculated area of the hair bulb The percentage of the total area of the hair bulb and the image area ratio is defined as the area of the hair bulb. The abrasion resistance is analyzed by calculating the relative value of the area of the hair bulb. The calculation method of the maximum hair bulb area This has been extracted area of the hair bulb, can be obtained here as largest area of the hair bulb. Extracting a feature value of the result (after running) Extracting a feature value the results of running by the above method are shown in the following table, Standard sample according to the repeated shots, the last adjustment, came up with Table 1. Table 1 Eigenvalue extraction results(the standard sample photo of plain fabric) eigenvalue hair bulb hair bulb area the maximum area of numbe(n) percentage the hair bulb level 1 >=23 >=0.72 >=

5 Applied Mechanics and Materials Vols Improved Minimum Distance Discrimination Linear discriminant function, non-linear discriminant function, the minimum distance classification cluster classification discriminant function are three main method. The weighting coefficients of these two methods of the linear discriminant function, and non-linear discriminant function to select the more important, also more complicated, since the meaning of each element of the feature vector of the image of the extracted ball sample each independently, minimum distance classifier, is unknown characteristic parameter values of the image and the standard image, compared to find the minimum distance from the two, and the corresponding standard image is most similar to the unknown image, thereby to determine the position image attribution. Find the unknown category vector the unknown category vector attributable to the minimum distance a discriminant method to identify each of the categories on behalf of the vector of the center point of the distance. the minimum distance discrimination method to discriminate principle as follows: Ui = [uil Ui2 uid],m class: Ll, L2..., Lm, each class has a set of vectors as the most standard feature vector. For any one of the identification image, X = [Xl X2 ], calculating the distance d (X, U0), if there exists an i, so that d (X, Ui) <d (x, 1.1i) i J, j = 1,2,, m, i.e. Li recently to a template, then X is Li. Hair ball in the ball in the image characteristic parameter values extracted from the actual fabric, the normalization process and create a vector[3]. This approach to implement the programming is more complex. Improved Minimum Distance Discrimination is used to determine the level of fabiric wear reesistance According to the three characteristic values of the area and the maximum number of hair bulb of hair bulb, using the modified minimum distance discrimination law pilling rating. Combine experience with computer artificial assessed. The formula is: MINi=k1(x1-Ui1) 2 +k2(x2-ui2) 2 +k3(x3-ui3) 2 (1) Where k1, k2,... the manual to determine the right weight, may be adjusted according to actual situation, Ui1, Ui2, and Ui3 are the characteristic parameters of the standard sample of one level. According to three characteristic parameters of the number of hair bulb mainly selected area percentage of the hair bulb, hair ball maximum area. average eigenvalues is selected to be Ui. where the minumun of MINi is judged,the level of the fabric wear resistance is determined.processing results are shown in Fig.9. Fig.9 The image process software interface

6 426 Information Technology Applications in Industry Summary In this paper Matlab software is used to program software of image processing. Pilling fabric five Level standard sample according to image a series of processing and analysis, to verify and select the actual fabric; prove feasible, accurate results to improve Pilling the ball image processing data determine the minimum distance method for multi-feature image judged feasible, simplified minimum distance method to judge the program design, the experiment proved to be feasible. Acknowledgements This work was financially supported by 2011ZK-07of Clothing Materials Research Development and of Evaluation of Beijing Key Laboratory and Beijing Education Commission, Textile testing technology based on machine vision research project of BIFT university-level academic innovation team 2011T-06 College Talents school plan - the backbone of the talents of young and middle-aged project (PHR ). References [1] S.P. Gao, Li-Ming Wang. Signal Processing in MATLAB for fabric pilling images Preliminary Qingdao University, School of Computer Science thesis,(2007) [2]H.K.Lu,Zh.L.Zhong. Wavelet analysis theory in fabric Pilling the ball image denoising applications. Jiangsu Textile A, (2006),p.39 [3] Cao Fei. Based on image analysis technology fabric pilling grading method.(2007) [4] Das, A, Chakroborty, P.Study of Fabric Hairiness using Image Processing Technique.Research Journal of Textile & Apparel, (2008), p.562 [5] F.Hu. wool fabric pilling research and analysis. Wuhan University of Technology, (2002 ),p.109 [6] Y.Y Zhou, R. Pan, Gao Weidong. given standard sample according to the image analysis of fabric pilling ball equality method. Textile Technology Progress Report, (2010 ),p.29

7 Information Technology Applications in Industry / Improved Minimum Distance Discrimination Method Used in Image Analysis of Fabric Wear Resistance /