A Novel Framework for Real and Fake Smile Detection from Videos

Abstract

Smile is and has always been an evident parameter for judging one's state of mind. An indicator of emotions, a smile can be categorized into two types. Some are real, originating from an exhilarated atmosphere, while some are fake. Hence, it becomes utterly difficult to differentiate between the two smiles. This research work is based on capturing the movement of zygomatic major and obicularis oculli which plays a vital role in detecting whether a smile is fake or real. The appearance of wrinkles on the cheeks, corner of the mouth, indicate the contraction of the zygomatic major muscle, whereas the eye elongation indicates the obicularis oculli contraction. We have primarily worked on Videos in which the main emphasis is on the images of facial parts such as lips, eyes and cheeks area to distinguish between real and fake smile. The requisite portion of frames of training data videos are extracted and GIST is applied to it which is further trained by SVM. For test videos, the nature of frames for each video is predicted and based on the majority of real or fake frames in a video, it is classified into fake or real. Results show that the best accuracy in detecting true and fake smiles is close to 76.66%, while in reality, human true-fake-smile recognition ability is much lower. Thus, our work assures efficient output which could be used as a tool for the analysis of smiles in the psychological area and this research work can be further extended to detect fake and real expressions.