Fig. 1. Monash shape measurement system arrangement.

Fig. 12. 3D surface model of the 020117b measurement set. Fig. 13. Accurate segmentation for comparison to Fig. 15. Watershed of the manually specified segmentation identifying all of the surface rock fragments in the 020117b measurement set. Fig. 14. Calculated seed regions for the 020117b measurement set. Fig. 15. Calculated segmentation, approximating the visible fragments for the 020117b measurement set.

Fig. 2. Image overlay on the 3D surface model for the 020117b measurement set.

Fig. 3. 3D coordinate data of the central large fragment for the 020117b measurement set.

Fig. 4. Data filtered out of the 020117b measurement set using a 1 mm radius median filter. The absolute value of the largest 95% of z values is shown.

Fig. 5. Edges of occluded areas for the 020117b measurement set.

Fig. 6. Edge detection using the black top hat operator with a 10 mm radius structuring element and 5% threshold on z depth on the 020117b measurement set.

Fig. 7. Detected fragment edges for the 020117b measurement set.

Fig. 8. Distance transform applied to the final edges data shown in Fig. 7 showing the distance to the nearest fragment edge.

Fig. 9. Local maxima of the distance transform data set shown in Fig. 8 showing the distance to the nearest fragment edge.

Fig. 10. Creating seed region R_a by merging seed circles built up from distance transform local maxima points a, b, and c.

Fig. 11. Fragment edges detected using the black top hat morphological operation using a 3 mm radius spherical structuring element. Applied to the median filtered 020117b measurement set.

Experimental rock pile size classes and proportions

Counts of visible surface rocks by sieve size

Execution times for the base morphological operations on measurement 020117a

Execution times for the segmentation of measurement 020117b

Number of matching, under-segmented and over-segmented regions between the “accurate” and calculated segmentations

mch, Number of correctly matching regions, for which there is only one primary match between the “accurate” and calculated regions; useg, number of under-segmented regions in the calculated segmentation; oseg, number of regions in the “accurate” segmentation that are over-segmented in the calculated segmentation; % err, number of under-segmented and over-segmented regions as a percentage of the total number of regions.