Reproducible Surface Extraction for Variance Comparison in 3D Computed Tomography

This paper describes a novel method for creating surfaces models from distorted volume datasets in 3D computed tomography (3D-CT). As all 3D-CT datasets are prone to artefacts, especially geometry extraction may produce erroneous surface models if a single global threshold is used. Depending on the selected threshold either the area of material is thickened because ambient noise is added (threshold too low) or it is thinned because outlying material regions are classified as air (threshold too high).We propose a pipeline model which creates a reproducible output using common 3D image processing filters: First of all we use an edge preserving diffusion filter to reduce noise without blurring the edges of the specimen. Furthermore, a watershed segmentation filter is applied to the gradient image in order to extract a binary volume of the dataset. In the final step the surface model is constructed using an advanced approach called elastic surface nets. The major contribution of this paper is the development of the specific processing pipeline for extracting surface models of homogeneous industrial components and to handle large resolution data of industrial CT scanners. The pipeline is crucial for the following visual inspection of deviations.