Automated microscopy image analysis of sintered cBN materials

Two protocols for multistep grain segmentation and analysis workflow in optical microscopy images of cubic boron nitride materials were developed and compared. One is based on statistical region merging and second one on morphological segmentation of grains without high contrast borders. Judging from corresponding manual image segmentation by expert, the second method gave more accurate grain boundaries and better statistical correspondence. Then, using the morphological segmentation method and incorporating of parameter optimization into it, a grain analysis workflow was established. Deviations from the correct answer (expert segmentation) were quantified based on five geometric statistical indices, and these deviations were added together to define the overall error. Cross-validation confirmed that the morphological segmentation workflow reproduces the expert segmentation with smaller 9.4% margin of error compared to 23.9% with statistical region merging one. The automated grain segmentation of such challenging materials with high throughput image analysis is an important help for industrial development of new milling tools.