Actor Recognition for Interactive Querying and Automatic Annotation in Digital Video

S. Verstockt, S. Van Leuven, R. Van de Walle, E. Dermaut, S. Torrelle, and W. Gevaert (Belgium)

Keywords

CBIR, video querying, video annotation, face recognition,face detection, frame buffering, Viola Jones, Gaborwavelets.

Abstract

Interactive video querying and automatic annotation of video sequences are two research topics in the computer vision domain that are still in an early stage of development. Yet, no algorithms exist that are capable of adequately bridging the semantic gap in content based image retrieval and the automatic creation of video metadata is still far from ideal. The actor recognition algorithm proposed in this paper must be seen as a first step towards a solution for the general object recognition problem encountered in both domains. Our technique automatically detects and recognizes actors in digital video using intelligent frame buffering with object tracking, Viola Jones-based face detection, and Gabor wavelets-based face recognition. Using the interactive click-and-search technology, a simple click on the actor is sufficient to trigger the software to detect the face and to search in a known database of actors. Extensive experimentation of the interactive video querying on different video sequences achieved a retrieval efficiency of approximately 75% for a top 3-hit. Since the variance in pose and orientation of the faces, and the illumination differences in the queries are very high, this result is more than acceptable.

Important Links:



Go Back