Texture classification using kernel independent component analysis

We propose a novel method, kernel independent component analysis (KICA), for texture features extraction. The texture images are first mapped into a higher-dimensional implicit feature space. Then a set of nonlinear basis functions are learned using KICA. The feature vectors are obtained by projected the texture images onto the basis functions. Comparison experiments between KICA and the other two classic methods: Gabor filters and ICA, are performed. The results indicate that the KICA is an efficient approach for texture classification.