Automatic Recognition And Classification Of Woven Fabric Based On Image Processing

In the textile industry, the classification of woven fabric always depends on manual work, that wastes time and manpower. Meanwhile, the efficiency and reliability of inspection are limited due to fatigue and inattentiveness. Therefore, it’s necessary to propose an automatic and efficient classification system for the recognition and classification of woven fabric. The automatic model not only speed up the production, but also liberate large quantities of labour force. This paper intrduces the image processing and image analysis into the texture classification to achieve the automatic classification of the woven fabric types. There are many kinds of fabrics, such as woven fabric, knitted fabric and non-woven fabric. This paper takes woven fabrics as experimental objects, which include plain weave, twill weave and satin weave. The decision rules for recognizing warp and weft floats are developed, based on geometric features of yarn distribution.This paper proposes an approach to extract image features for woven fabric recognition. In the feature extraction phase, the uniform local binary pattern method is adopted to compute the occurrence frequencies of all the rotation invariant patterns in the proximity of local pixels when R=1,2,3with P=8,16,24. Meanwhile, it combines with Gabor wavelet is utilized to calculate the mean and standard deviation of each filtered image as features. Facing so many features, it employs principal component analysis to reduce the dimension of features. At last, an appropriate classifier of the support vector machine is used as a classifier in the classification phase. The experimental result indicates that the uniform local binary pattern and Gabor wavelet method can automatically and accurately classify the woven fabric types, and the recognition rate of woven fabric ups to93.33%.Then, this paper proposes another automatic woven fabrics image extraction and classification method. Firstly, two-dimensional wavelet transform is utilized to decompose the preprocessing image, in order to gain the1/16size of the low-frequency sub image for fabric image Secondly, extracting the feature parameters of gray level co-occurrence matrix of the low-frequency sub image:energy, contrast, correlation and entropy. Lastly, the support vector machine is used to establish and train the neural network to classify three mean weave:plain weave, twill weave, stain weave. The experiment results demonstrat that gray level co-occurrence matrix extracts the features well, and the support vector machine can be recognized three basic woven fabric structures automatically and accurately. The recognition rate of woven fabric is86.66%.With the development of image processing technology, the recognition and classification of woven fabric will turn from the manual work to automatic model quickly.