Automatic Face Annotation in Images
Automatic image annotation has been rapidly gaining attention in recent years. Among many objects of interest, the annotation of human faces has a number of potential applications of broad interest, e.g. family album organization and character-based search of commercial films or news videos. In this work, we propose a face annotation framework for real-world applications.
To achieve greater accuracy in recognizing faces under various viewing conditions, local distinctive invariant features—namely Scale-Invariant Feature Transform (SIFT) features, which were originally designed for reliable matching between different views of the same object—are incorporated into this framework. Our experiments demonstrate that the bag-of-SIFT representation serves as a good face description, especially under conditions in which facial expressions and poses vary drastically.
