Identification of Cashmere/Wool Based on Pairwise Rotation Invariant Co-occurrence Local Binary Pattern

Abstract

The identification of cashmere and wool fibers is a challenge of textile field. The two animal fibers are very similar in surface morphology, performance of physics and chemistry. In this paper, we proposed a new method for automatic identification of cashmere and wool fibers with high accuracy. The Pairwise Rotation Invariant Co-occurrence Local Binary Patterns was used to represent the microscopic images of cashmere/wool fiber. Every fiber image was converted to a vector, which is a histogram of LBPs extracted from fiber images. The vectors were fed into Support Vector Machine for a supervised classification. The experimental results indicated that identification accuracy is about 90% and the proposed method is robust under datasets with various blend ratios.