NCJ Number
243277
Date Published
2013
Length
8 pages
Annotation
This project developed, evaluated, and documented a system for facial verification that uses detected facial-marks rather than holistic or larger scale facial structures.
Abstract
The project developed a robust facial-mark detection method that provides both location and scale information for facial marks. The project has shown the utility of facial marks in face-image retrieval tasks using a large-scale face-image database. The proposed facial-mark matching scheme can be combined with COTS face matcher to improve the overall matching accuracy. The improvements in face matching accuracy are shown both on the general face-image database and the identical-twin face-image database. The project then proposed a conditional filtering scheme in order to reduce the retrieval time on a large-scale face-image database. The study of symbolic query base face-image retrieval showed a great utility for facial marks when query image is not available. The project concludes that although a number cases were found in which facial marks can be used for face-image retrieval tasks, the instability of facial marks is still a major challenge in using facial marks in practice. The instability of facial marks is mainly caused by the low quality of face mug-shot images. The development of a more robust facial-mark detection and representation scheme and establishing a standard practice in face mug-shot image capture that would ensure sufficient quality of image are some of the imminent requirements for extending the use of facial marks. 3 figures