I am looking for the performance of state of the art methods for face recognition of a person with different emotions or expressions.
check the paper Local Directional Number Pattern and other related local patterns for emotion and expression recognition
There is s very good software for that.
http://www.noldus.com/human-behavior-research/products/facereader
it works even with a webcam (7000 dollars though)
Although the question seems simple, actually it is not.
When you would look at accuracies reported by commercial software producers (like Noldus or others) you would get rates about 95% or even 98%. What they don't tell, is how they obtain them - usually it is the best case scenario tested in steril environments and for two-class emotions classification.
Accuracy depends on: lightning condition, camera resolution, position of face towards the camera, face atypical overriding with: glasses, moustache, extra coiffures etc. as well as subject's ability to express emotional states asked or evoked.
In many experimental settings highest accuracies are about 75%-80% and that even drops down in in-the-wild condition.
Hope this helps...
- Badges
- Science topic