Attribute classification for the analysis of genuineness of facial expressions

FERNANDEZ FLORIO, GONZALEZ; MARIA ELENA BUEMI; DANIEL ACEVEDO; NEGRI, PABLO

Curicó

Conferencia; International Conference on Pattern Recognition Systems; 2021

In this work we study many different machine learning based approaches (particularly, artificial neural networks variants) to classify instances of facial expressions on video according to its genuineness. Comparing to most problems that computers learned to solve using artificial intelligence, this problem is a task not trivial to solve by human beings. With that comes the difficulty of evaluating the performance of the developed models.We use the SASE-FE dataset that was designed specifically to solve this particular problem. This dataset contains videos of subjects doing facial expressions, labelled with type of expression and truth value (if it is genuine or not).The main analysis is based on comparing deep feed-forward neural networks with recurrent neural networks. This particular type of network is known for being capable of extracting information from a sequence and keep it through time. In that way, a video can be classified using not only its features but also the ones from its predecessors.Since the amount of videos in the dataset is not abundant, a new metric is proposed to make a more granular analysis and that allows to compare with more detail the results that each implemented variant gives. Results suggest that certain facial features that allows distinguishing a genuine expression and a faked one are too related to the subject that performs them and then, developing an universal classifier (independent of the subject) seems unfeasible.Regarding the comparison between the two types of networks, although the recurrent variants cannot overcome the values obtained by the deep variants, we can appreciate that they reach similar results but with a smaller amount of training epochs.