INDUSTRIAL AND COMMERCIAL APPLICATION Automatic classification of seam pucker images based on ordinal quality grades Ioannis G. Mariolis • Evangelos S. Dermatas Received: 15 August 2010 / Accepted: 14 September 2011 / Published online: 4 October 2011 Ó Springer-Verlag London Limited 2011 Abstract Seam pucker evaluation is based on the anno- tation of seam specimens into discrete grades of quality presenting an ordinal arrangement. Therefore, an ordinal logistic regression (OLR) model is highly qualified to fit the features extracted by such specimens into the assigned quality grades. This work focuses on building an automatic system for seam pucker evaluation based on an OLR model and comparing its performance with a similar system that employs an ordinary least squares (OLS) regression model. In this direction, three separate types of features have been extracted by a dataset of 325 seam images. The OLR model outperformed OLS for all feature types showing that its theoretical advantage applies also in practice. The best OLR model produced a correct classification rate of 81.2%, matching up to the performance of human experts. Keywords Visual inspection  Seam pucker  Surface roughness  Ordinal logistic regression 1 Introduction Quality control is a fundamental process in textile and clothing manufacturing industry. Usually it is performed by human personnel and, therefore, is subjective, costly and time consuming. Seam puckering and fabric wrinkling are two examples where quality control is accomplished by human experts through visual inspection and comparison to certain quality standards. Seam puckering is highly corre- lated to seam quality and appears when the smoothness along the seam line is disturbed by the appearance of vertical waves. It is caused by uneven ply feeding, exces- sive thread tension or structural jamming. On the other hand, fabric wrinkling is an important property used in the evaluation of the esthetic and visual appearance of fabric surface. International Standards Organization has established ISO 7770 [1], regarding assessment of seam quality. It refers to seam specimens and uses standard images for assigning to each specimen a quality grade. In the sewing industry, seam specimens are used by experts on a trial basis, to determine the optimal settings for the sewing machine’s parameters. The French method of ‘‘cylindre creux.’’ applied to wrinkled fabrics also assigns quality grades, according to NF G 07-125 standard. Both ISO 7770 and NF G 07-125 standard employ five discrete quality grades, where grade 1 denotes the worse and grade 5 denotes the best seam or wrinkle quality, respectively. The aforementioned standards introduce some objec- tivity to the quality assessment. However, as long as the human judgment is involved, the procedure remains sub- jective and inefficient. Therefore, efforts are made toward automatic seam and wrinkle quality control. In previous work by Fan et al. [2], a laser scanner and Chebyshev band pass filtering have been employed to extract fabric surface profiles that are parallel to the seam line. Four statistical parameters produced by these profiles have been compared to subjective pucker grades, two of which presented excellent correlation. In Park and Kang [3], a method using five shape parameters for seam pucker evaluation has been proposed. It is employing three-dimensional image analysis and neurofuzzy engines to obtain quantitative evaluations of the seam pucker. A cognitive model has been developed in Stylios and Sotomi [4] regarding the measurement of seam pucker in lightweight synthetic fabrics. The model I. G. Mariolis (&)  E. S. Dermatas Department of Electrical and Computer Engineering, University of Patras, 26500 Patras, Greece e-mail: yannismariolis@upatras.gr 123 Pattern Anal Applic (2013) 16:447–457 DOI 10.1007/s10044-011-0241-y