A new set of texture features based on the Haar transform
Tor Lønnestad · 2003
A new set of features for textural classification of images is presented. The features have a natural ordering in frequency and direction, and the size of the feature set can easily be adapted to each application. Basing the method on the Haar transform makes it computationally efficient. The classification performance is compared to that of gray level cooccurrence matrix (GLCM) features and gray level run length matrix (GLRLM) features, and the Haar features have shown they classify better than GLCM and GLRLM features, using both clustering and supervised minimum distance classification.>