Fabric Defects Detection Research Based On Image Processing Technology

Fabric defect detection is a research focus in the field of computer vision, and there have been a lot of research results. Because the current detecting technology can't fully meet the needs of actual production from the point of effectiveness, real-time performance and applicability, so it is necessary to explore new technology of fabric defect detection. After conducting research on many current algorithms for fabric defect detection, three new ones are proposed in the thesis.Firstly, the traditional local binary pattern operator, LBP, is improved by introducing standard deviation statistics. The improved operator considers not only the gray value relationship of pixels in the local neighborhood, but also the relationship between the mean square error of whole image and that of the local neighborhood pixels. The fabric defect detection algorithm based on the improved LBP operator is developed. Experimental results show its defect detection rate is higher than the traditional one.Secondly, ideal band-pass filter and Gaussian band-pass filter are designed on the basis of the detailed analysis of fabric image's frequency spectrum. The center frequency's setting problem of relevant filter for variant textures fabric image is discussed. And two types of filter banks are constructed in order to detect fabric defects according to the principle that adjacent filters of the bank are tangent at the half-peak of their amplitude response. Experimental results prove that this detection algorithm can effectively detect defects of different texture images.Lastly, three types of Log-Gabor filter banks with multi-scale and multi-orientation are designed based on fabric image's frequency spectrum and the principle that adjacent filters of the bank are tangent at the half-peak of their amplitude response. The fabric defect detection algorithm using Log-Gabor filter bank has relatively better detection results.