Computationally Efficient Invariant Facial Expression Recognition

The most important bottleneck for facial expression recognition system is recognizing the expression in uncontrolled environments with minimum computational time consumption. This problem has been addressed by combining the robust local texture descriptors which are invariant to illumination effects. In this work, the illumination effects are eliminated by using Weber Local Descriptor (WLD). Next, Local Ternary Descriptor (LTP) was introduced to preserved discriminatory local information, which is robust to noise. Both types of features are concatenated to produce more discriminatory feature set. The proposed technique is computationally efficient and gives very good results on JAFFE face database.