Is it a good task for a classifier? Result with too many false positives.

asked 2016-10-13 06:56:07 -0500

These last days I've been working on training a classifier to find some patterns on images. Specifically, those red and blue auras in the next image (however they can take another random color):

image description

For this task I decided to train a classifier. Finding a lot of these auras (500) in different images and using them as positive samples. I have tested several types of classifiers, however I'm obtaining a lot of false positive results.

I wonder if doing this with a classifier is a good task, due these auras have a color which emphasizes them (but, again, it's not always the same color) and, perhaps, it's a task for image processing techniques. Isn't it?

Could you give me some guides to perform this detection?

edit retag flag offensive close merge delete


i don't think, you get far with training cascades on this, they won't ever look at colors, you only got an "outer" shape (and no further texture), also those shapes are not "rigid"

berak gravatar imageberak ( 2016-10-14 00:15:08 -0500 )edit