Tomography is an important technique for non-invasive imaging, with applications in medicine, materials research
and industry. Tomographic reconstructions are typically gray-scale images, that can possibly contain a wide
spectrum of grey levels. Segmentation of these grey level images is an important step to obtain quantitative
information from tomographic datasets. Thresholding schemes are often used in practice, as they are easy to
implement and use. However, if the tomogram exhibits variations in the intensity throughout the image, it is not
possible to obtain an accurate segmentation using a single, global threshold. Instead, local thresholding schemes
can be applied that use a varying threshold, depending on local characteristics of the tomogram. Selecting the
best local thresholds is not a straightforward task, as local image features (such as the local histogram) often do
not provide sufficient information for choosing a proper threshold.
In this paper, we propose a new criterion for selecting local thresholds, based on the available projection data,
from which the tomogram was initially computed. By reprojecting the segmented image, a comparison can be
made with the measured projection data. This yields a quantitative measure of the quality of the segmentation.
By minimizing the difference between the computed and measured projections, optimal local thresholds can be
Simulation experiments have been performed, comparing the result of our local thresholding approach with
global thresholding. Our results demonstrate that the local thresholding approach yields segmentations that are
significantly more accurate, in particular when the tomogram contains artifacts.