Recognition and analysis of fabric texture by double-sided fusion of transmission and reflection images under compound light source

Abstract

Computer vision is widely used in fabric texture recognition. In this paper, a new method based on double-sided fusion of reflection image and transmission image is proposed for recognition and analysis of fabric texture. The yarn location is obtained through transmission image, and fabric texture is obtained through reflection image. The position information of weave floats obtained by gray projection on the transmission image is given to the reflection image for classification and color recognition of weave floats. In the stage of weave float classification, an improved KNN algorithm based on FCM is proposed. First, FCM is used to cluster the HOG features of the two types of floats respectively to obtain new cluster centers, and then KNN is used for classification to obtain the weave patterns. In the stage of color recognition, the K-means clustering algorithm is used on a single weave float to obtain the color pattern. Based on the two attributes of a single point obtained from the above two patterns, a system of bidirectional error correction is designed. Experimental results show that this method effectively improves the recognition accuracy of yarn-dyed fabrics which interlaced yarns are different colors without color and texture disturbed to the greatest extent.