Defect Detection in Fabrics Using Local Binary Patterns - Advances in Image and Graphics Technologies - page 282

the increase of P, the running time of the program is also increasing. In other words,
LBP 24,3 consumes a longer time than LBP 16,2 , which is a little slower than LBP 8,1 . From
the overall view, the detection results of employing LBP operator for the fabric of the
delicate texture (e.g. defects A-E) are more obvious than for the fabric of the coarse
texture. With the increasing of roughness of fabric texture, the testing effect will be the
less desirable (e.g. defects F-H). In addition, the LBP algorithm is also available to the
patterned fabrics (e.g. defect I-J).

5

Summary

LBP is proposed as a method to detect the fabric defects in this paper. The relative gray
value between pixels is considered as response in LBP algorithm. Thus it is invariant
towards the monotonous gray changing. The original LBP operator is improved to make
it own rotation invariance and set the threshold to determine the uniform patterns. As a
result, the algorithm can effectively reduce the dimensions of feature values and the
computation time, because only the information of uniform patterns is analyzed. Make
the similarity of the feature vectors of both the detection windows and the whole image
as the criterion to confirm defects. In this way defects can be identified and segmented
more accurately. Experimental results have shown that the detection results of LBP
algorithm are reliable whether in the side of intuitive visual or in terms of the false
detection rate. However, LBP algorithm is not available to detect all kinds of defects on
all texture backgrounds. Consequently it still needs a further improvement in the area of
fabric defect detection to accommodate different fabric texture backgrounds.

Acknowledgement. The authors gratefully thank the Scientific Research Program
Funded by National Natural Science Foundation of China (61301276), Shaanxi
Provincial Education Department (Program No. 2013JK1084), Shaanxi Science and
Technology Research and Development Project (Project No.2013K07-32).

References

1. Li, L.Q., et al.: Image Processing Progress in Fabric Defect Automatic Detection. Donghua
University (Natural Science), 李立轻等 图像处理用于织物疵点自动检测的研究进展 .
东华大学学报 ( 自然科学版 ) 28(4), 118-122 (2002)
2. Cho, C.S., Chung, B.M., Park, M.J.: Development of real-time vision-based fabric inspec-
tion system. IEEE Transactions on Industrial Electronics 52(4), 1073-1079 (2005)
3. Kumar, A.: Computer-Vision-Based fabric defect Detection: A Survey. IEEE Transactions
on Industrial Electronics 55(1), 348-363 (2008)
4. TSai, I.-S., Lin, C.-H., Lin, J.-J.: Applying an Artificial Neural Network to Pattern Recog-
nition in Fabric Defects. Textile Research Journal 65(3), 123-130 (1995)
5. Manjunath, B.S., Chellappa, R.: Unsupervised texture segmentation using Markov Random
Filed Models. IEEE Transactions on Pattern Analysis 13(5), 478-482 (1991)