Can we estimate the perceived comfort of virtual human faces using visual cues?

摘要

The sense of strangeness (or discomfort) perceived in certain virtual characters, discussed in Uncanny Valey (UV) theory, can be a key factor in our perceptual and cognitive discrimination. Understanding how this strangeness happens is essential to avoid it in the process of modeling virtual humans. In this paper, we investigate the relationship between images features and the discomfort that human beings can perceive. We extract image features based on Hu Moments (Hum) and Histogram Oriented Gradient (Hog). The saliency detection is also extracted in the specific parts of the virtual face. Finally, a model using Support Vector Machine (SVM) to provide binary classification is suggested. The results indicate accuracy of around 80% in the image estimation process comparing with subjective classification. As a contribution, some areas may benefit from this study for avoiding the creation of characters that may cause strangeness, such as the games, conversational agents and cinema industry.