The main purpose of the project is recognition of emotions based on facial expressions. Cohn-Kanade data set (http://www.pitt.edu/~emotion/ck-spread.htm) is used for explorations and training.
The Cohn-Kanade is a data set of face images labeled for 7 expressions/emotions :
- Anger
- Contempt
- Disgust
- Fear
- Happiness
- Sadness
- Surprise
The solution consists of versatile classifiers including Decision Tree, Random Forests, Support Vector Machines, DT based Binary Classifier, simple hybrid classifier.
Performance results, estimated by cross validation tests:
- Multi-class Support Vector Machines (radial kernel, kernlab): 97.3% (st. deviation = 0.4%)
- Multi-class Support Vector Machines (linear kernel, e1071): 96.5%
- Decision Tree based classifier: 89.63%