Extraction of fibres in tomographic samples of composites by computer vision
Résumé
Tomography and other 3D imaging technologies produce large volumes of data for even moderately sized samples, discouraging exhaustive manual analysis and repeated measures. However, relative limited resolution, low dynamic range, quantization levels, and noise make difficult the automated analysis of the samples. Thus, exploiting at full length the wealth of information contained in a tomography becomes a challenging task. We present a study case of extraction of fibres from a synchrotron microtomographic image of a carbon/carbon composite, allowing the detailed characterization of the material. We detail the steps to extract the fibres from a sample having all the typical drawbacks, and give some comparative data for labelling algorithms for 3D matrices. The classical recursive and two pass iterative techniques, together with a hybrid approach, are compared.