Robust facial expression recognition based on median ternary pattern (MTP)

Facial expression is a strong medium to express one´s feelings and emotions. Accurate detection of facial expression can convey a lot of information about a person´s mood. The major barriers to accurate recognition are the presence of noise, illumination variation and occlusion. This paper proposes an effective appearance-based facial feature descriptor constructed with a new local texture pattern, namely the median ternary pattern (MTP) for facial expression recognition. The proposed MTP operator encodes the texture information of a local neighborhood by thresholding against a local median gray-scale value and quantizing the intensity values of the neighborhood around each pixel into three different levels. The MTP codes generated for an image or image patch is then used as the feature representation of the facial expression image. The effectiveness of the proposed method has been evaluated using images from the Cohn-Kanade (CK) Expression Database. Classification was done using Support Vector Machine (SVM). Experimental results show increased accuracy in recognition rates when using the proposed approach in comparison to other popular gray-scale based methods.