The aim of this study is to simulate the pareidolia capability of humans to produce an emotional response to a scene using analysis of facial expressions associated with abstract face-like patterns. We developed a system that uses a holistic face detector and a facial expression classifier. The υ and SVDD One-Class Support Vector Machines (SVM) were evaluated for creating a holistic face detector, which looks for faces that can vary from natural faces to minimal face-like patterns. A Pairwise Adaptive C and υ-SVM (pa-SVM) were evaluated for creating the facial expression classifier. In both scenarios, a dataset of human faces and facial expressions was used to produce a number of preprocessed images (grayscale, histogram equalised grayscale; and their respective Sobel and Canny edges) at a number of resolutions for analysis. A Gaussian and a degree two polynomial kernel were used with the SVM methods and the results were obtained using a 10 fold cross validation technique. A concern with the face detectors is verifying that they can look for minimal face-like patterns empirically. To address this concern, we created cartoon faces of the human face dataset and degraded these cartoon faces to produce an array of minimal face-like patterns. We then evaluated the face detectors and facial expression classifiers with the best model parameters on these cartoon faces. The outcome is a holistic system with the potential to describe a scene by producing an array of emotion scores corresponding to Ekman's seven Universal Facial Expressions of Emotion.
2013 IEEE Symposium on Computational Intelligence for Creativity and Affective Computing (CICAC). Proceedings of 2013 IEEE Symposium on Computational Intelligence for Creativity and Affective Computing (CICAC) (Singapore 16-19 April, 2013) p. 79-86