印象情報付きPCAによる日本人とタイ人間の人種的顔特徴の抽出

概要

論文の詳細を見る
We used facial images of racially close group such as Japanese and Thais to study which features affect the viewers impression of ethnicity. Principle component analysis (PCA) was done to extract the racial features of the sample group. The facial features were separated into information on texture and shape, and the PCA was applied to each. We compared the basic with the weighted PCA with the score related to impression of race. The racial impression score was provided by a psychological experiment. A weighted PCA with information of the users impression extracted racial facial features from an image that had fewer cues to ethnicity because of changed facial direction and reduced lighting. The extracted features of texture and shape could be used for the racial impression conversion of facial images. We suggest that the identification of features typical of a given race relies on detailed texture rather than shape.