Adaptive image classification with radial basis function network

This paper proposes an adaptive image classification scheme, characterized by the use of rotation-invariant polar-wavelet energy signatures, a reduced feature vector of dominant energy signatures for each texture image, and a trained Radial Basis Function (RBF) network classifier. The classification of a given texture image involves computing the polar-wavelet energy signatures, adaptive selection of dominant energy signatures to form a reduced feature vector, and feeding the feature vector to the trained RBF neural network for texture classification. The experimental results, based on testing 7200 rotated texture images, show that the proposed classification method is effective for rotation invariant image classification.