Defect Detection in Fabrics Using Local Binary Patterns - Advances in Image and Graphics Technologies - page 275

study on fabric defect detection with the help of Gauss - Markov random field (GMRF)
model. This method is not restricted by the types of defects, however, shortcomings
also exist. The amount of calculating the estimated model parameters is very large and
time-consuming is quite long. The methods based on spectrum are suitable for partic-
ular textured fabric. Wavelet transform [6], [7], [8], Gabor filters [9] as well as the
discrete Fourier transform (DFT) [10], [11] are typical spectrum based methods of
fabric defect detection. With such methods, the defect parts of the fabric image are
highlighted mainly by time-frequency analysis and detected through the next thre-
sholding. Nevertheless the time-frequency transform operations will reduce the detec-
tion speed, and has a large amount of calculation. As a result, it is difficult to apply such
methods to detection online directly.
A method based on LBP is proposed in this paper. The global and local feature
values of fabric images are extracted by LBP and further analyzed. Then the similarity
can be calculated with Chi-square function. Set a threshold to determine the defect area.
Thus, the purpose of defect detection is achieved.

2 Local Binary Patterns

LBP is originally put forward by Ojala [12] in 1999. For the reasons that LBP can be
regarded as an effective texture description operator and its prominent ability of de-
scribing the local texture features of the images, LBP has been widely used in the area
of the description of image texture.
The basic idea of LBP algorithm is built on pixels. Comparing the gray values of the
center pixel with its surrounding neighborhood pixels, relative gray can be obtained and
considered as the response of the center pixel. Therefore, LBP is invariable for mo-
notonic gray-scale changes. Basic LBP operator is shown in Figure 1.

Binary: 110110001 = Decimal: 177

Fig. 1. Basic LBP operator

In this model, a neighborhood of the 3×3 window is regarded as the processing unit.
The gray value of the center pixel is regarded as the threshold. The gray values of the
surrounding eight neighbors pixels are compared with the threshold and processed by
binarization. If the gray value of the pixel is smaller than the center pixel gray value, it