Fuzzy clustering of multispectral images based on integrated spectral and spatial features

Abstract

An unsupervised classification method in multispectral images based on a fuzzy clustering driven by integrated spectral and spatial features is presented. Spectral information can be obtained directly from pixel values in different frequency-band images, while spatial information can be extracted by mean of texture analysis. The images are spectrally decorrelated via the Karhunen-Loeve Transform (KLT). The resulting first principal component image is then exploited by applying the gray-level co-occurrence matrices. A fuzzy clustering approach is finally performed based on the features that combine both spectral and spatial information on the image. The results of applying the algorithm to an urban area image are illustrated.