Partial automation of database acquisition in the FAVRET face tracking and recognition system using a bootstrap approach

摘要

The FAVRET face recognition system uses a database built from multiple labeled views of each individual to achieve both profile-to-profile tracking of face pose and pose-invariant recognition. Adding a new individual to the database, however, is non-trivial. This work describes the use of a modified version of the FAVRET system as a module in a graphical user interface (GUI) for database acquisition, which automatically detects and tracks face regions in input video or still-frame material and estimates the data which would otherwise have to be entered manually. The estimates can be adjusted manually if necessary. The first application is to update the database in the GUI module itself, using data more accurate than that from which the original prototype database was built.