S. Mukaida, H. Ando, K. Kinoshita, M. Kamachi, and K. Chihara (Japan)
Facial image, Age, PCA, Shape feature, Image synthesis
This paper presents a statistical method that can extract age features from face images for synthesizing novel age manipulated face images. We propose a new statisti cal feature extraction method that extends standard PCA (Principal Component Analysis) by adding a known at tribute such as an age to input variables. This method can derive the first principal component that is strongly related to the given attribute. An experiment was con ducted to test the proposed feature extraction method, and we confirmed that extracted age-related shape features were quantitatively and qualitatively sufficient. We also synthesized a number of age-manipulated face im ages using this method, and conducted an age evaluation experiment to test how human subjects estimate age of the synthesized faces. The result showed that estimated age increased as the manipulated age increased, which indicates that the extracted age features are visually effective. The proposed feature extraction method is solely based on facial shapes.
Important Links:
Go Back
