0

I am trying to identify set of colors in bottles which are kept serially using OpenCV. I have tried the following procedures but most of them fail in different colors of ambient light.

  1. Image segmentation using Histogram but fails for ambient light condition
  2. Image segmentation using K-means but that also fails for ambient light colors.
  3. Tried the color masking technique by defining the colors I am looking for but that isn't too great too.
  4. White balancing the image and then applying the above techniques. But that technique gives problematic colors normal white color background.

Any recommendation on this would be much appreciated. A reference image is attached.

enter image description here

I am looking for the following result. I would like to detect light yellow, light yellow, peach, peach, orange, orange, orange, orange, ivory, pink. enter image description here

user2101977
  • 113
  • 1
  • 9
  • I have a bunch of questions: What is the purpose? Identifying bottles based on the content? Tracking some kind of reaction over time? Are the colors predefined / a limited group or simply the detected color? Why is the 3rd bottle labeled '03' a lighter shade of orange? It seems pretty similar to the 2nd '03'. I suppose detecting the label is not an option? What exactly did you try for white balance? Can you also provide an image in a different / difficult ambient light? – J.D. Oct 08 '21 at 10:19
  • break it down. get an instance segmentation. that's the hard part. the rest is just poking pixels. – Christoph Rackwitz Oct 08 '21 at 16:17

0 Answers0