Abstract
Caricatures sometimes impress people more than original photographs. This paper aims to improve the automatic caricature system in [1] by generating more expressive caricatures based on analyzing the features of face images with impressions of ethnicity, gender, and expressions. An eigen space model is set up by the principal component analysis of 2D edge information of faces. Qualitative and quantitative description of differences and importance of principal components of each impression are given by the linear discriminant analysis method. By using Support Vector Machine method, the mixed eigen space is divided into four, that is, Japanese male/female, Caucasian male/female. Towards Japanese male input images, expressions transformations by learning transformation rules between six basic expressions and exaggerated caricatures by the caricaturist are carried out. More expressive caricatures can be drawn and animated by the proposed method.
