Stylized Textile Image Pattern Classification Using SIFT Keypoint Histograms

摘要

Semantic image classification, which is the process of categorizing images using pattern recognition technology, is very useful for image annotation, organization and retrieval. While the literature has focused on the classification of natural scene photographs or images, here we focus on the stylized textile images and this is totally a new area which is in the domain of artificial images. In this paper, we show that SIFT keypoint histograms perform much better than the traditional gray level co-occurrence matrix with the SVM classifier. Furthermore, we create a low-dimensional representation for each image using principle component analysis (PCA) method to the SIFT keypoint histograms and achieve a better result. To the best of our knowledge, this is the first time the SIFT feature histograms has been used to the classification of stylized textile images.