Indian Semi-Acted Facial Expression (iSAFE) Dataset for Human Emotions Recognition

Abstract

Human emotion recognition is an imperative step to handle human computer interactions. It supports several machine learning based applications, including IoT cloud societal applications such as smart driving or smart living applications or medical applications. In fact, the dataset relating to human emotions remains as a crucial pre-requisite for designing efficient machine learning algorithms or applications. The traditionally available datasets are not specific to the Indian context, which lead to an arduous task for designing efficient region-specific applications. In this paper, we propose a new dataset that reveals the human emotions that are specific to India. The proposed dataset was developed at the IoT Cloud Research Laboratory of IIIT-Kottayam – the dataset contains 395 clips of 44 volunteers between 17 to 22 years of age; face expressions were captured when volunteers were asked to watch a few stimulant videos; the facial expressions were self annotated by the volunteers and they were cross annotated by annotators. In addition, the developed dataset was analyzed using ResNet34 neural network and the baseline of the dataset was provided for future research and developments in the human computer interaction domain.