A multi-scale framework to segment the neurons.
A mathematically rigorous general approach for the normalization of the response of a multi-scale ensemble of linear filters.
A multi-scale framework to compute the Laplacian of the 3D image stack and an approach to compute as many decision functions as the number of scales (one for each scale) used for segmentation.
A mathematical justification for using different low-pass filters to compute the Laplacian and the Hessian matrix.
An extensive experimental evaluation of the performance of our approach on a number of datasets, including all of the DIADEM competition.