Investigating the temporal association between eye actions and smiles

With the significant use of facial expressions in conveying information in human to human interaction, it has become quite imperative to be able to evaluate expressions especially for security reasons. Humans have grown to master ways of exhibiting misleading facial expressions and this has become part of our life style, maybe to avoid being rude or to hide malicious intent. This problem may be alleviated by checking temporal relationships between basic facial actions as people exhibit expressions. This paper seeks to find the temporal relationship between eye actions and smiles. Once patterns are noted, a standard framework can be formulated to authenticate expressions based on those trends. An AAM-SVM based training model is proposed to recognise smiles and blinks respectively. Onset and offset times of both expressions are noted and analysed to study the correlation between smiles and blinks. We have chosen two posed and spontaneous datasets which are the Extended Cohn-Kanade (CK+) and the Denver intensity of spontaneous facial action (DISFA) databases, respectively. A strong correlation between blinks and smile onsets for blinks that occur before smile onset is observed on posed smiles, while a strong correlation is observed on blinks that occur towards the end of a smile in natural and spontaneous smiles.