Image Segmentation with Tensor–Based Classification of N–Point Correlation Functions

In this paper, we investigate the use of N–Point correlation functions from material science literature, for medical image segmentation, and introduce a classifier suitable for use with these functions. The N–point correlation functions serve as good estimators of component material distributions and their packing in a multi–phase heterogeneous media. We represent these multi–phase properties with tensor structures and employ these functions as features in our tensor decomposition based classification algorithm. We use a variant of Higher Order Singular Value Decomposition (HOSVD) to extract the multi– linear properties of the tensor feature space and reduce the dimensionality with respect to several modes. The preliminary results of segmenting a placenta image with these functions and classifier, are very promising.