Legacy ImageMagick Discussions Archive

I picked my threshold of 4% just based upon a good visual threshold of the one yes_90a.jpg image. I have not experimented either.

I think a larger test/training set will be needed to pin down the best threshold value or to see if either method holds up (or what percent failure one gets on a larger test set).

Yes.

I notice that the two methods, although they both distinguish between "yes" and "no", have different orders of images within each set. This suggests that either method might have difficulties with a larger set of images.