Texture Segmentation Using Independent Component Analysis of Gabor Features

This paper proposes a novel method for texture segmentation using independent component analysis (ICA) of Gabor features (called ICAG). It has three distinguished aspects: (1) Gabor wavelets transformation first produces distinct textural features characterized by spatial locality, scale and orientation selectivity; (2) principal component analysis (PCA) then reduces the dimensionality of these features and ICA finally derives independent features for texture segmentation; and (3) two different frameworks for ICA are discussed. Framework I regards pixels as random variables and represents them as a column vector by re-shaping all the transformed images row-by-row, while framework II treats the statistical features, viz. the mean and standard deviation of image, as random variables. The statistical features of all the transformed images construct a column vector. Comparative experiment results among ICAG, Gabor wavelets and ICA indicate that ICAG provides the best performance and framework II is more efficient and applicable for texture segmentation