Supervised texture classification by integration of multiple texture methods and evaluation windows

Pixel-based texture classifiers and segmenters typically combine texture feature extraction methods belonging to a same family. Each method is evaluated over square windows of the same size, which is chosen experimentally. This paper proposes a pixel-based texture classifier that integrates multiple texture feature extraction methods from different families, with each method being evaluated over multiple windows of different size. Experimental results show that this integration scheme leads to significantly better results than well-known supervised and unsupervised texture classifiers based on specific families of texture methods. A practical application to fabric defect detection is also presented.