How to build your own eye detection classifier algorithm?

asked 2015-03-28 19:16:38 -0500

PeterWeter gravatar image

I created an eye classifier with 278 self-cut eye-pictures and 31 random background pictures. It detects eyes 96% of the time but also a lot of false positives. How to approach this task?

  1. Build a stronger classifier, ie more pictures?
  2. Do that later in software? How?
  3. ...
edit retag flag offensive close merge delete

Comments

I've never trained a classifier, but a 10:1 ration doesn't sound right. What happens if you take more negative samples?

FooBar gravatar imageFooBar ( 2015-03-29 07:29:02 -0500 )edit

Yes... typically having twice as much negative examples as there are positive ones is a good place to start.

Pedro Batista gravatar imagePedro Batista ( 2015-03-30 11:17:11 -0500 )edit