Module: Find_Sides

Description:

Labels edge pixels of a roughly axis-aligned object, a label corresponding to one of the six sides X-, X+, Y-, Y+, Z-, and Z+. Those labels are respectively 10, 20, 30, 40, 50, and 60.

As a side effect, edge pixels not belonging to the outer boundary of the object are also labelled, their label is 1. All other pixels are labelled 0.

Algorithm

1. Use a line-of-sight approach from each side of the image border to flag edge pixels according to the side and direction of the line. Flags are not mutually exclusive.
2. All edge pixels with only 1 flag are marked A, those with 0 or more than 1 flag are marked B.
3. Perform a region-growing from A into B, using specified connectivity.
4. Apply a given number of passes of a sort of majority filter on A: for each pixel in A, if the clear majority of its neighbourhood (including itself) is of a different flag, change it to the majority.
5. For the flags corresponding to each of the six sides, find the biggest connected component, all other are marked B.
6. Perform a region-growing from A into B again.
7. Label edge pixels according to rule described above.

Illustration on a brick-like object with some surface roughness. Result drawn with the Draw_Cubes module.

Notes

• A pixel can belong only to one side, excluding 1-pixel thick objects.
• The algorithm was built for object with six contiguous sides. Staircase-like objects are not considered.
• Step 3 of the algorithm removes thin like branches that can snake from one label into another. In practice, a one-pass clean looks alright, but it depends on the object shape.

Connections:

Image

Ports:

Threshold

Comparison

Connectivity

Illustration of the different connectivities. (a): 6-, (b): 18-, and (c): 26-connectivity (in case you didn't figure it out, n-connectivity means n pixels in the neighbourhood).

Cleaning

Compute

Commands:

verbose

create