Keyword and face image retrieval based on latent semantic indexing

To annotate images in keywords and to map visual features to symbolic features are critical problems in sense of multimedia information processing. This work describes mechanisms for retrieving suitable keywords on face images and for retrieving face images based on latent semantic indexing in a face image annotation and a retrieval system. The visual features are represented in sizes/lengths of face parts, and symbolic features in keywords, respectively. A space is constructed by applying latent semantic indexing, which is called the latent semantic space. This space consists of face images and keywords. Also, this space reflects the sizes/lengths of the face pans. Retrieving face images and keywords in terms of sizes/lengths of face parts, and retrieving keywords, which are suitable to be assigned to the face images, are achieved. Retrieving keywords using the sizes/lengths is corresponding to assignment of keywords to one face image.