Article Preview
TopIntroduction
Facial expressions provide a key mechanism for understanding and conveying emotions. It has been suggested that the ability of humans to interpret emotions is very important for effective communication, accounting for up to 93% of communication used in a normal conversation (Yaffe, 2011). Emotion recognition is highly desirable in future machines for the ideal Human-Computer Interface (HCI) (Ko, 2018). Computer applications could better communicate if they were able to change their responses according to the emotional state of users in various interactions. By creating machines that can understand emotion, the communication that exists between humans and computers can be enhanced. This would open a variety of possibilities in robotics and HCI such as devices that warn a drowsy driver, attempt to placate an angry customer, or better meet user needs in general.
Emotion Recognition is a difficult task that requires skilled people to do it manually. However, this can be automated with the help of Machine Learning. Recent advances in Machine Learning have helped to recognize emotions efficiently and effectively by decreasing the cost of manual labor with increased convenience.
This work proposes a framework for emotion recognition based on facial expressions in still images with the help of Machine Learning. Machine Learning is the study that gives computers the ability to learn without being explicitly programmed (Koza J.R., 1996). It is generally categorized into three types: Supervised Learning, Unsupervised Learning and Reinforcement Learning. In Supervised Learning, the machine experiences input data that is labeled for each example which helps the algorithm to correlate the features (Stuart J. Russell, 2010). Whereas in Unsupervised Learning, the input data is not labeled (Hinton & Sejnowski, 1999). Reinforcement learning refers to a computational approach of understanding and automating goal-directed learning and decision-making to attain a complex objective (Barto, 2015). Two of the most common supervised machine learning tasks are regression and classification. In regression problems, the machine predicts the value of a continuous response variable (Rencher & Christensen, 2012). Whereas in classification problems the machine learns to predict discrete values, i.e., the most probable category, class, or label for new examples (Har-Peled, 2003).
The given data can be converted to useful information by following standard procedures. The process of finding relevant materials to satisfy an information need is known as information retrieval. For example, the picture of a person can be used to detect their emotional state. Here, the picture is the data and the process of associating emotion with that picture requires extracting information out of the data and processing it. It may be noted that the initial dataset cannot be used for classification as some of the features in the input data may be irrelevant or noisy. Appropriate extraction methods are used for image information retrieval (Goodrum, 2000). Information processing is used in diverse fields and applications. This paper proposes some pipelines to recognize the emotional state of a person using a still image of their face. This processing of information is not just a computationally involved task but is also useful for society. Therefore, suitable feature extraction algorithms are used to build derived features intended to be informative and non-redundant, facilitating the subsequent learning and generalization steps. Feature extraction can also help in dimensionality reduction leading to better human interpretations (Alpaydin, 2010). To further improve this, feature selection algorithms have been used to extract a relevant subset of features. Feature Selection is the process of selecting a subset of relevant features wherein a feature is a characteristic of the phenomenon being observed. There are two types of feature selection, univariate and multivariate (James, Witten, Hastie, & Tibshirani, 2013).