Face Image Annotation Based on Latent Semantic Space and Rules

摘要

This paper presents a face image annotation system based on latent semantic indexing and rules. To achieve annotation, visual and symbolic features are integrated. Two features are corresponding to lengths and/or widths of face parts and keywords, respectively. In order to develop annotation mechanism, it is required to vary the dimensions of the spaces which are constructed by the latent semantic indexing, and to represent direct relationships among features. Associated symbolic features to visual features are represented in rules based on decision trees. Co-occurrence relationships among keywords are represented in associa-tion rules.