MULTI-RESOLUTION IMAGE SEGMENTATION USING THE 2-POINT CORRELATION FUNCTIONS

Recently, the 2-point correlation functions (2-pcfs) were employed in building feature vectors for histological image segmentation. The 2-pcfs serve as estimators of material distributions with respect to the component packing in a multi-phase sample. The multi-phase properties estimated by the 2-pcfs were represented in a tensor structure and a HOSVD-based classification algorithm was developed. In this paper, we employ a multi-resolution framework in the image and the 2-pcfs feature scale-space, in order to achieve significant savings in computational costs. We also propose a new formulation of the HOSVD classifier that learns the relative skew in the feature space. The classifier helps in improving the segmentation accuracy. Our improved results are validated against ground-truth generated from large histology images of mouse placenta.