This study proposes a novel evaluation method for quantifying perceived translucency using statistical facial image analysis. Most conventional evaluation methods for impressions, including perceived translucency, typically use the skin of a certain “part” of the face as representative for evaluation. However, we anticipate that most impressions such as perceived translucency would be recalled from features of the face as a whole, or from various parts of the face (not from one specific area). Therefore, a novel evaluation method that targets images of the entire face should be proposed instead of the conventional methods. In order to develop such a facial image analysis, we employed an image analysis based on principal component analysis (PCA) that is useful for analyzing textural features of the entire face. To begin with, as the learning data necessary for “statistically” extracting the facial textural features, we created a database of shape-normalized facial image data where all of the facial shapes were normalized (unified) and only the facial textures were different. Next, we obtained the eigenspaces that express the textural features of faces using PCA. As the last step, by using the characteristics of eigenspaces, we developed a novel “Eigen residual accumulation method” that would allow us to quantify textural features of faces. The comparative evaluation between the proposed method and subjective assessment for the perceived translucency of faces showed that the method quantifies the level of the translucency with high accuracy.
View full abstract