Rotation-invariant texture classification using a two-stage waveletpacket feature approach

摘要

A novel two-stage wavelet packet feature approach forclassification of rotated textured images is discussed. In the firststage, a set of sorted and dominant wavelet packet features is extractedfrom a texture image and a Mahalanobis distance classifier is employedto output N best classes. In the second stage, another set of waveletpacket features is extracted from the polarised form of the sampletexture image and the most dominant wavelet packet features are selectedand passed to the radial basis function (RBF) classifier with the N bestclasses to output the final matched class. Experimental results, basedon a large sample data set of twenty distinct natural textures selectedfrom the Brodatz album with different orientations, show that theproposed method outperforms the similar wavelet methods and the otherrotation invariant texture classification schemes, and an overallaccuracy rate of 91.4% was achieved