Three-dimensional boundary following

Abstract A graph search based method for boundary following in a stack of serial section images is presented. The cost function utilizes edge gradient magnitudes, edge circularity, and comparisons with neighboring sections. The performance of the method is compared for 2- and 3-dimensional gradient operators and for cost functions which omit the circularity component and/or comparisons. Results suggest that the use of 3-dimensional gradient operators greatly reduces search effort, and somewhat improves the accuracy of boundaries for noisy images.