Classification of Low Level Visual Texture Features Based on the Hermite Transform

Alfonso Estudillo-Romero, Boris Escalante-Ramírez

Abstract: A biological inspired image analysis technique to extract visual texture features is presented. The Hermite transform describes locally basic image features in terms of Gaussian derivatives. Multiresolution combined with several derivative orders of analysis provides detection of patterns that characterize every texture class. Maximum energy direction analysis and steering of the transformation coefficients increase the method robustness against the texture orientation. Texture features are computed by extracting statistical information from the orientation-invariant visual features and arranged into a compact vector. The PCA technique is used to select the most significant linear combinations of the vector elements to reduce vector dimensionality. We evaluate the correct classification rate for several kinds of texture features with real textures sets and the effects of the number of principal components on the classification performance.