Texture classification using wavelet packet and Fourier transforms

A new texture classification algorithm using wavelet packet transform is proposed. It uses principal component analysis technique and statistical distance measurement to combine and select frequency channel features to give improved classification performance. Comparison is also made between wavelet packet transform features and Fourier transform features on a set of eight optical texture images with several level of white noise added. Both algorithms are successfully applied to the classification of under-ice sidescan sonar images.