Visualization of Relationship between Facial Features and Impression Evaluations

Impression of facial images strongly relates to the tones and shapes of the faces. This paper presents our study on the visualization of relationships between facial features and impression evaluations. Specifically, we generate clusters of faces and then visualize the relationships between the clusters and impression evaluation results. This study applies a set of artificial facial images generated with a machine learning model and then extracts facial regions. We apply a clustering algorithm with features of the faces calculated based on pixel values and aspect ratios of the faces. Meanwhile, we conduct impression evaluations based on the SD method on the five-point Likert scale. This study integrates the clustering and impression evaluation results and visualizes the dataset by using our own hierarchical data visualization technique.