2

I have a classification task on a number of gray-scale images (m classes and n images in each class). Before showing my classification results, I want to show that the intra-class variation between the images is high and the inter-class variation is low. This is apparent when we see the images but how do we formulate it into a mathematical result? There was a question on comparing two matrices on Comparing two matrices in Matlab but I want to do it on n matrices of a class.enter image description here

enter image description here

Could there be a criteria (class dependent) which could tell us that an image is x distance off the class average?

Community
  • 1
  • 1
Zafar
  • 23
  • 5
  • I get the general idea of what you're looking for, but showing a few images would make a world of difference. Some measurements would work great to show what you want for certain kinds of images; others would be awful. – John May 17 '13 at 05:35
  • The images are of human gestures. They have a lot of variations in rotation, scale and deformable silhouettes. – Zafar May 17 '13 at 05:42
  • I have added two images that belong to the same class. – Zafar May 17 '13 at 05:52

0 Answers0