Face Image Annotation in Impressive Words by Integrating Latent Semantic Spaces and Rules

摘要

This paper describes a mechanism to annotate face images in impressive words which express their visual impressions. An annotation mechanism is developed by integrating latent semantic indexing, decision trees, and association rules. Moreover, visual and symbolic features of faces are integrated, which are corresponding to lengths and/or widths of face parts and impressive words, respectively. Relationships among these features are represented in a latent semantic space, their direct relationships in decision trees, and co-occurrence relationships among symbolic features in association rules, respectively. Efficiency of annotation results is improved by integrating these mechanisms, since their features are utilized effectively.