COLOUR SEGMENTATION IN YARN-DYED WOVEN FABRIC IMAGES BY USING K- MEANS CLUSTERING Bekir Yildirim 1, Brigita Kolčavová Sirková 2

Transcription

1 COLOUR SEGMENTATION IN YARN-DYED WOVEN FABRIC IMAGES BY USING K- MEANS CLUSTERING Bekir Yildirim 1, Brigita Kolčavová Sirková 2 1 Faculty of Engineering, University of Erciyes, Turkey 2 Faculty of Textile Engineering, Technical University of Liberec, Czech Republic Corresponding author: Bekir Yildirim; bekiryildirim@erciyes.edu.tr Extended abstract Visual quality is one of the most important properties of textile fabrics which directly affect the value of the product and also the consumer desire to buy. Therefore, careful fabric inspection is very important for fabric manufacturers especially for today s highly competitive global textile market [1,2]. Colour is one of the most important quality parameters that should hit the target for sample production and also needs to be maintained through the whole production process. Colour control is realized by visual inspection in industry for multicolour woven fabrics where warp and weft yarns have different colours. This is mostly due to the fact that spectrophotometers or colorimeters which are used for measurements of solid colour woven fabrics where warp and weft yarns are same colour are not capable of measuring multicolour regions. Lately multi-spectral imaging systems have been researched for colour measurements which are capable of not only measuring colour but also the spatial information. While using these systems it is needed to separate multi-colour region and solid colour region automatically. Obtained segmentation information can further be used for evaluating colour defects of woven fabrics. The main aim of this study is to separate colours of yarns in digital images of multicolour woven fabrics acquired in RGB colour space for further processes. Colour Based Segmentation Acquired images are in RGB Colour space composed of 3 channels and they needed to be converted in to CIELAB colour space for colour segmentation due to the fact that in this colour space it is possible to exclude the lightness information and deal only with the colour properties. The L coordinate represents lightness, the (a) coordinate represents; (+ a) redness or ( a) greenness, and the b coordinate represents; (+b) yellowness or ( b) blueness. The colour information is included in 'a*' and 'b*' layers.

2 Classify the Colours in 'a*b*' Space Using K-Means Clustering The woven fabric images are decomposed into L, a* and b* colour channels and every pixel of the image contains three channel values. It is needed to determine which pixels represent the same colour by using this information. Clustering can be used in order to separate groups of objects represented by features. Due to the fact that colour information is stored in a* and b* colour channels, the colour segmentation turns to a clustering problem where the pixel colour values represent the features of the individual pixel locations. For this purpose k-means clustering can be used which partitions a data set (x1, x2,, xn) into k disjoint subsets S = {C1, C2,, Ck} such that a clustering criterion is optimized (Maulik & Bandyopadhyay, 2002). It finds partitions such that objects within each cluster are as close to each other as possible, and as far from objects in other clusters as possible. The most widely used clustering criterion is the sum of the squared Euclidean distances (Jain, Murty, & Flynnl, 1999) between each data point xj and the cluster centre μk of the cluster Ck. Using k-means algorithm, we need to specify the expected number of clusters k which is a positive integer number. Misclassified pixels are eliminated by using filtering methods after clustering. To examine the performance of the proposed method, different real woven fabric samples are used. The woven fabric samples including different weaves and different coloured yarns for patterned fabric samples, which were obtained from Liberec Technical University Weaving Laboratory, are used for trials. The specifications of the woven fabrics and yarns are as follows. Table 1. Specifications of the woven Fabrics and Yarns Warp Weft Sett 80/100 mm 150/100 mm Yarn Raw Material 100 % Acriylic 100 % CO Yarn Number (tex) 2x35 tex 5x30 tex The woven fabrics are scanned using a flatbed scanner in true colours in resolution of 50, 100, 150, 300, 600, 1000 and 1200 dpi. The images are acquired from both sides of the woven fabrics. While acquiring the image of the woven fabrics a black paper is placed at the back side of the fabric so as to obtain constant background colour for reflecting light. The size of woven fabric samples is 18cm x12cm. The proposed algorithm is carried out by writing a computer programme in MATLAB (R2013b). For illustration purposes a 300 x 300 (pixel) region is selected in 300 dpi images and implemented algorithm results are shown.

3 Conclusions It can be concluded from the study that the k-means clustering algorithm can be used for classifying pixels into different colour groups by means of colour information gathered from a* and b* channels. This segmentation can be used for obtaining colour information separately in a multi-colour woven fabric and can be used for further evaluation of colour defects. The method also brings about some drawbacks, particularly the need of cluster number determination before using the algorithm.

4 Acknowledgements The Scientific and Technological Research Council of Turkey (TUBITAK) is acknowledged for granting of B.Yıldırım postdoctoral study in the framework of TUBITAK-BIDEB 2219 International Postdoctoral Research Scholarship Programme. References 1. Xie X. A review of recent advances in surface defect detection using texture analysis techniques. Electronic Lett. on Comp Vision and Image Analysis, (2008). 7(3), Tunák, Maroš, Aleš Linka, and Petr Volf. Automatic assessing and monitoring of weaving density. Fibers and Polymers, 2009, 10(6), Çelik H.İ., Dülger L.C. & Topalbekiroglu M. Development of a machine vision system: real-time fabric defect detection and classification with neural networks, The Journal of the Textile Institute, (2014). 105(6), William K. Pratt, Digital Image Processing-PIKS Scientific Inside 4thed, A Wiley-Interscience publication. 5. Jain A.K., Murty M.N. & Flynn P.J. Data clustering: A review. ACM Computing Surveys, (1999). 31(3), Maulik U. & Bandyopadhyay S. Performance evaluation of some clustering algorithms and validity indices. IEEE Transactions on Pattern Analysis and Machine Intelligence, (2002). 24(12),

5 7.