Photometric Stereo-based Woven Fabric Pattern Recognition Using Wavelet Image Scattering

The weave pattern is a crucial factor that enhances the strength and stability of the fabric. Pattern recognition of woven fabric based on vision methods has been widely developed. In this research, woven fabric's basic weaving pattern recognition is based on photometric stereo images. First, six images of woven fabric were taken, each with a different direction of light. Next, an unbiased photometric stereo algorithm was used to reconstruct the six images. This paper used 23 grayscale photometric stereo images measuring 400 x 300 pixels. Augmentation techniques were carried out to produce 458 images consisting of 240 plain woven images, 159 twill woven images, and 60 satin woven images. The training data set consists of 367 images, and testing consists of 192 images. The feature extraction method uses wavelet image scattering and classification using Principal Component Analysis (PCA) and Support Vector Machine (SVM). The wavelet image scattering method effectively extracts texture features of photometric stereo images of diverse woven fabrics, while the PCA and SVM methods successfully classify the basic woven fabric patterns. The results of recognizing the basic woven fabric pattern using PCA and SVM classification obtained an accuracy of 98.57%.