Routine Metallography: Machine Learning Segmentation, the Future of Image Analysis
ZEISS Microscopy
Abstract
Image analysis is a well-established technique for extracting information from images, particularly but not limited to microstructures captured with light microscopy. Digital image segmentation, via gray level thresholding, histogram-based or edge detection techniques, has become well accepted in grain size and phase analysis.
Automated segmentation is reliable but not always an exact science, and patterns in images that may be evident to the human eye may not be identifiable by computer, and vice versa. Machine learning is able to combine human and computer insight, and has particular value in analyzing sets of images, even complex multiphase materials images, and large datasets require intuitive and fast training workflows. This webinar presents the basis, user interface and results of machine learning segmentations of metal and material samples, which expand the image segmentation possibilities of the analyst. In doing so, more precise segmentation data is able to generate more enlightening insight.
Key Learnings:
- Understand the limitations of traditional image segmentation
- Learn how ZEISS solutions enable intuitive and fast training of machine learning models
- Explore case studies where machine learning is able to improve upon traditional segmentation in metals and materials