3D differential Canny Edge Detector

Differential Canny Edge Detector for volumetric three-dimensional data

This package contains a 3D differential Canny edge detector (edgedetect.m) and a simple segmentation routine (segmentation.m) that uses the calculated edges to segment the data.

The Canny edge detector consists of three steps:
1) De-noise the data using a Gaussian convolution filter.
2) Get candidates for the edge set by calculating maxima of the image gradient magnitude. This is performed by finding the zero-level-set of a certain function (which contains second derivatives of the data).
3) Perform hysteresis thresholding. First, remove all parts of the edge set with gradient magnitude (edge strength) below the lower threshold. Then, remove all connected components of the edge set where the gradient magnitude is never over the upper threshold.

The segmentation routine is a wrapper for the image processing toolbox function bwlabeln. Using the computed edge set, it assigns all voxels in the data space either 0 (edge) or 1 (non-edge), increases the edge surface thickness (to close possible holes) and then calls bwlabeln to segment the data.

For some mathematical details, see Appendix B of http://arxiv.org/abs/1403.2620