Modelling of Plain Weave Fabric Structure and Its Use in Fabric Defect Identification

Texture is an inherent property of a fabric whose periodicity can be extracted from DC Suppressed Fourier Power Spectrum Sum (DCSFPSS). Periodicity can be used as one of the fabric quality parameters to detect the woven fabric defects. This paper focuses on modeling periodicity of a plain weave fabric based on DCSFPSS and using this model to detect the fabric defects. The nonparametric and parametric modeling were experimented on 1-D DCSFSS data as a signal where the effectiveness of parametric method in modeling normal fabric was evident. Parametric methods viz., Autoregressive (AR), and Autoregressive Moving Average (ARMA) models were tested on DCSFPSS of a normal fabric image. Performance parameter viz., %fit function for u direction of DCSFPSS was found to be 97.1%/93.8/95.4 for ARMA (64,64) / ARMA (32,32) /AR (32) indicating superiority of ARMA (64,64) over others but found to be computationally complex. For fabric defect detection, a comparatively simple AR (32) model showed for u/v direction of DCSFPSS, a fit function spread of 5%/5% for normal sample against that of ~43%/62% for loose weft defect sample. These facts justify that, a simple AR (32) models well the periodicity of the fabric for u/v direction of DCSFPSS and conclusively differentiates defective fabric from normal plain fabric samples.