0

I am writing genetic program to extract object in image. I have two fitness functions for now: Hamming and Hausdorff distance.

Hausdorff is good but too slow for bigger images. Hamming is faster but not always works (for examples sometimes white image is the best in population). Do you know any other fitness function which contains advandtages of this two functions.

This is binary images. Third way is to first find edges or resize images and then use Hamming distance to evaluate solution and for now it is the best way.

  • What images do you compare? Original image of object on background and..? Candidate region on solid/gradient background filled with average color of the rest of image? Or blurred rest of an image? – Yuri Gor Oct 07 '17 at 20:40
  • @YuriGor I used Hamming function which is count pixels on diff(xor) of two images and Hausdorf functions which measures how far two subsets of a metric space are from each other. You can read more about it on wikipedia. My images has only black and white pixels. Could you tell me where I can read about others method, which you mentioned? Not only for binary images, but also for greyscale images. –  Oct 12 '17 at 20:19

1 Answers1

0

NORM_L1 is very fast in OpenCV and works well for GA. (I used it for grayscale images, but I think it will work fine for 3 channels too)

Yuri Gor
  • 1,353
  • 12
  • 26