A new method for the detection and classification of weave pattern repeat

Abstract

A novel method combining the characteristics of corner information and second-order statistics of fabric images is proposed for analyzing the repeat size and identifying the weave type of fabrics. The analysis method is formulated in a research framework of near regular texture analysis. Firstly, the original fabric image is split into two components: the geometric component and the textural component, in which low-level cues of corners are detected by the Harris detector. Secondly, clustering techniques are used to cluster interest points by image patch appearance. Thirdly, a Markov Random Field (MRF) model is adopted for inferring the location of texture elements, and their shape is then used for classifying the weave patterns. The experimental results show that the main disadvantages and difficulties of using structural and frequency methods in texture classifications have been overcome by the proposed method.