1

I am working on counting colonies. After the great deal of effort, I can finally get the binary images. But I have no idea about how to split the overlapped ellipses.

Here are the original images

img

and binary images

img

Hope to get the tips from experienced. Thank you.

Spektre
  • 49,595
  • 11
  • 110
  • 380
ITVision
  • 21
  • 1
  • Try to use [Circular Hough Transform](https://en.wikipedia.org/wiki/Circle_Hough_Transform). – Didgeridoo May 15 '16 at 20:18
  • Welcome to stackoverflow. You are supposed to show some own effort before others help answering your question. Make sure you have read: http://stackoverflow.com/help/how-to-ask – tfv May 16 '16 at 06:20
  • However, your question has a lot of interesting aspects. Can you share the coding you have used to generate binary images? This is not a simple threshold, but you have somehow compensated for ununiform illumination. In a first step, I'd modify that threshold so that colony sizes are smaller, some of thsoe colonies will fall apart by this (in the original image you'll see that borders between colonies are slightly darker). – tfv May 16 '16 at 06:23
  • First of all, I'd also try to make sure that illumination is uniform and that also the background is uniform (you can see shades there) – tfv May 16 '16 at 06:28
  • Possible duplicate of [Algorithms: Ellipse matching](http://stackoverflow.com/questions/36042396/algorithms-ellipse-matching) – Spektre May 16 '16 at 09:32
  • 1
    see the duplicate **QA** but there are also different approaches like count the colonies total surface area in pixels and divide by average colony area size. **BTW** colony count is not a good parameter better is the area itself but in times where this was measured without computers and cameras it was more practical to count colonies instead (unless this is used for measuring the average density of micro organism somewhere). – Spektre May 16 '16 at 09:35
  • 1
    I'm sorry, I didn't fully understand the question. You can also measure the invariant characteristics for each black shape such as [shape compactness factor](https://en.wikipedia.org/wiki/Shape_factor_(image_analysis_and_microscopy)) `C`: `C = P^2 / A`, where `P` - perimeter, `A` - area. The most compactness shape is a circle: `C = 4*pi`. Also it is possible to find [image descrete central moments](https://en.wikipedia.org/wiki/Image_moment) and [shape elongation](http://link.springer.com/article/10.1007%2Fs10851-007-0039-0). If you want I may write the answer with more details. – Didgeridoo May 17 '16 at 22:15
  • @Didgeridoo I agree (+1) but for white shapes not black (probably just typo).... dismising the petry dish of coarse (by area or size) – Spektre May 18 '16 at 08:05

0 Answers0