Rotation invariant texture classification using directional filter bank and support vector machine

This paper presents a rotation invariant texture classification method using a special directional filter bank (DFB) and support vector machine (SVM). This method extracts a set of coefficient vectors from directional subband domain, and models them as multivariate Gaussian densities. Eigen-analysis is then applied to the covariance metrics of these density functions to form rotation invariant feature vectors. Classification is based on SVM, which only takes non-rotated images for training and uses images at various rotation angles for testing. Experimental results have shown that this DKB is very effective in capturing directional information of texture images, and the proposed rotation invariant feature generation and SVM classification method can in fact achieve relatively consistent classification accuracy on both non-rotated and rotated images.