Figure 3

Comparison between original and segmented data. Original data (a, e, i). Segmentation with incorrect (b, f, j) and correct (c, g, k) threshold values. Segmentation results (d, h, l). Top row: To separate a lipid droplet from a nucleus, a 2D slice of the original image (a) is segmented through a contour resulted by an incorrect (b) and a correct θ _min value (c), finally showing the segmented image (d). The scale bars indicate 5 μm. Middle row: 3D segmentation of a metaphase nucleus (e) with (f) and without (g) free histone attachments and a 3D mesh isosurface representation of the segmented slices (h). The dimensions of the bounding cube are 7.1 μm × 8.8 μm × 13.6 μm (e-g), 10.5 μm × 12.6 μm × 7.8 μm (h). Bottom row: Low contrast image stack (i) segmentation leads to adding visually significant false positive voxels by using a global threshold (8-bit threshold value θ = 44) (j). It is segmented correctly by 3D object-based segmentation (k), resulting in a labelled stack (l). The regular cubes in the left side of the panels (i)-(l) indicate 50 μm ³.