Transforming Pixel Signatures into an Improved Metric Space

Our previous work in computer-aided mammography has used scale orientation pixel signatures to provide a rich description of local structure. However, when treated as vectors for statistical classi cation, the Euclidean space they de ne has unsatisfactory metric properties. In this paper we describe a scheme that makes use of a previously described signature similarity measure to de ne a non-linear transformation of pixel signatures into a space with improved metric properties. K-means clustering is used to select prototype signatures and multidimensional scaling uses inter-prototype distances to form a new improved metric space. Finally, support vector regression is used to learn the transformation between the original and new spaces.