Comparison and fusion of multiresolution features for texture classification

We investigate the texture classification problem with individual and combined multiresolution features, i.e., dyadic wavelet frame, Gabor wavelet, and steerable pyramid. The support vector machines are used as classifiers. The experimental results show that the steerable pyramid and Gabor wavelet classify texture images with the highest accuracy, the wavelet frame follows them, and the dyadic wavelet significantly lags them. Experimental results on fused features demonstrate the combination of two feature sets always outperform each method individually. And the fused feature sets of multi-orientation decompositions and stationary wavelet achieve the highest accuracy.