Unsupervised Segmentation of Texture Images Using a Combination of Gabor and Wavelet Features

This paper presents an unsupervised method of texture classification by combining the two most commonly used multi-resolution, multi-channel filters: Gabor filters and wavelet transform. We used a set of 8 Gabor filters and 2 wavelet filters: Daubeschies and Haar, for our analysis. The parameters (viz. frequency, orientation and size) of the Gabor filter bank are obtained by trial and error method, based on visual observation of an energy measure of the response. A fuzzy classifier has been used which uses no a priori knowledge of the textures and hence provides unsupervised segmentation. For comparing the performance of the features from the Gabor filter bank, the two wavelet filters separately and a combination of all the three, the classification algorithm was kept identical. A combination of Gabor and wavelet features provides better performance compared to the individual features alone.