This paper proposes the template-based method to generate expressive caricatures automatically. This method consists of two steps, that is, the first step to extract and analyze facial parts in an input facial image, and the second step to select an appropriate template for each facial part. Facial parts such as eyes, eyebrows, nose and mouth are extracted from an input face image by using a shape-based pattern matching method. Feature parameters are calculated using principal component analysis on luminance distribution of facial parts extracted from the facial image database of 300 people. Next a hierarchical clustering method is applied to classify facial parts into multiple classes with similar shapes. Multi-class classifier is designed by SVM based on feature parameters and classified results. A class of each facial part in an input face is determined by the multi-class classifier and a caricature can be generated by assigning an appropriate template to each of facial parts. Each template represents the typical or exaggerated shape of selected class. The proposed method reduces the operator's load to interactively select and operate templates. Subjective evaluation of automatically generated caricatures shows the usefulness of the proposed method.
