Decoding facial expressions using a new normalized similarity index

The detection of mixed emotions from facial expressions is explored in this paper, in addition to the conventional categorization into six basic human emotions. A new distance-based normalized similarity index is defined for this purpose involving all the training templates in the normalization. The scale-, rotation- and illumination-invariant Difference Theoretic Texture Features (DTTF) are the image cues used for the classification. Experimentation on a benchmark facial expression dataset with comparisons to state of the art methods confirms the efficiency of our approach.