Are cascades the best way to detect different mouth shapes on photos?

edit

Hello,

I currently need to recognize different mouth shapes on photos (thin or thick lips, big or small mouth...). For now I have trained cascades for each mouth shape with about 30 photos of mouthes each (up to 1500 positive samples created with opencv_createsamples). The results are not convincing, the cascade for thin lips detects big lips and and vice versa.

So i wonder if cascades are the good way to do what I want or do I only need more positives samples?

I will also need to recognize the shapes of other face elements like eyes, eyebrows, nose... I am totally new to this technology, so has anyone already tried to do that? And how?

Thank you, Régis.