Feature Extraction Using Auto-Regression Spectral Analysis for Fabric Defect Detection
A new feature extraction method for fabric defect detection is proposed, which is based on one-dimensional projection series of fabric images. By using horizontal projection and vertical projection of the image, the characteristics of periodicity and orientation of fabric texture can be fully utilized. In terms of detection defects, it helps acquire information at most, and the computational complexity can also be greatly decreased with one-dimensional projection series. The proposed new method, named Auto-Regressive spectral analysis (AR), is a kind of modern spectral analysis method which is very suitable for analyzing short data with a high spectral resolution. The Burg algorithm is applied to estimate the AR spectrum. Finally, t-test is applied to verify the effectiveness of AR spectral features. This approach has been applied to various cases of defect detections with satisfactory results.
Lun Bai and Guo-Qiang Chen
J. Zhou et al., "Feature Extraction Using Auto-Regression Spectral Analysis for Fabric Defect Detection", Advanced Materials Research, Vols. 175-176, pp. 366-370, 2011