0

I have a bunch of skin images with moles. But unfortunately a large part of the set is containing annoying colored circles which I want to detect & remove. I've tried my best with cv2.HoughCircles but the results are completely wrong :(

Does anybody have an idea to detect these more efficiently? I've attached some images and my output.

Image 1 Image 2 My detection

zoma
  • 2,433
  • 1
  • 14
  • 24
  • 1
    What if you did some kind of analysis based on hue (in HSV or similar colour space). Those circles are definitely not skin coloured, so you ought to be able to segment them out reasonably well. | Could you show your current code, please, in some minimal variant -- just so that we can at least see what exactly you did and possibly reproduce it. – Dan Mašek May 03 '18 at 23:44

1 Answers1

4

If you just want to judge whether there exist colored patches, you can try use Saturation in HSV as this: How to detect colored patches in an image using OpenCV?.

  1. Convert to HSV from BGR
  2. Threshold on Saturation channel
  3. Other steps...

enter image description here

enter image description here

Kinght 金
  • 14,440
  • 4
  • 49
  • 62