~ 1 min read

Image Processing - Connected Component Analysis

Ahh, theres a result.

I’ve been working some image processing coursework trying to correctly label an B+W image with a different label for each object in the image…it’s taken ages!….

The basic premise is, go through each pixel in the image. If this pixel has a neighbour which is labelled, label it with that label, if two pixel neighbours labels differ, store the equivalence in a table and resolve it later.

Resolve the equivalence table (find the lowest equivalent neighbour and remember it). Do another pass to check if we still have equivalent items in the image. I’ve been struggling forever trying to find a way to resolve the equivalence table right down to the lowest level straight away in one pass, which is extremely difficult, so you perform multiple passes!

Anyway, glad that bits over – now onto the complex stats of it all!