DEVELOPMENT AND IMPLEMENTATION OF GRAIN BOUNDARY IDENTIFICATION ALGORITHMS

1 ZABA Krzysztof
Co-authors:
1 PUCHLERSKA Sandra 2 PYZIK Jaroslaw 1 HOJNY Marcin
Institutions:
1 AGH University of Science and Technology, Krakow, Poland, EU
2 Sabre Poland, Krakow, Poland, EU
Conference:
26th International Conference on Metallurgy and Materials, Hotel Voronez I, Brno, Czech Republic, EU, May 24th - 26th 2017
Proceedings:
Proceedings 26th International Conference on Metallurgy and Materials
Pages:
210-215
ISBN:
978-80-87294-79-6
ISSN:
2694-9296
Published:
9th January 2018
Proceedings of the conference were published in Web of Science and Scopus.
Metrics:
43 views / 16 downloads
Abstract

Current approach for microstructure images recognition and image-based properties measurement utilized manual iterative image filtering and threshold-binarization process until image areas, corresponding to desired structure elements, were extracted. This approach, however, led to information loss and inaccurate results, due to extensive filtering required to remove noise and extract features. Results were very sensitive to scanning method used to obtain images, image quality and coloring. Also, manual binarization process assumes that desired structure features are already known.In this article, we present and describe implementation and results of new approach, where image is segmented using algorithms based on Watershed [1], Morphological Geodesic Active Contours (MorphGAC), and Morphological Active Contours without Edges (MorphACWE) [2-3] algorithms, providing context-independent image partitioning. After image is segmented, obtained segments are classified and then measurements are taken for desired classes. This approach allows to find more features than binarization approach with higher accuracy, as minimal filtering is required, and MorphGAC/ACWE algorithms tend to be more accurate in edge and contour detection than simple thresholding or linear filters.Our program is written in Python, with use of OpenCV and Scikit-image libraries. It implements mentioned algorithm and provides tools for image filtering, and also analysis and measurements tools including features size and distribution statistics are available. For optimization enhancements, Python C-extensions and OpenCL-based GPU processing will be used if needed.Future enhancements include graph theory structure analysis, as image partitioned into segments corresponding to structure elements can be easily represented in a graph form. Our goal is also to utilize neural network for microstructure recognition, segmentation and property analysis.

Keywords: microstructure images recognition, grain boundary, images recognition algorithms, Python
Scroll to Top