Supervised Building Extraction Using Morphological Profiles with Adaptive Structures

Morphological profiles are one of the highly effective tools for image classification when structural information is critical. Morphological profiles, however create high dimensional feature space and increase the complexity of classifier. In this paper we have quantitatively evaluated the fact that the classification accuracy of each profile is dependent on the size and shape of the structuring element. We propose a classification scheme which uses morphological profiles with adaptive structuring element. We relate the shape and size of structuring element which is used for producing morphological profiles with the discrete wavelet transform of the image. The size and shape of structuring element adapts to the frequency content of the pixel's neighborhood. With this adaptive structuring element we can produce a single profile which is quite effective for classification. The results show that with proposed scheme significant improvement was obtained in the classification accuracy with reduced dimensionality of feature space.