Publications Repository - Gdańsk University of Technology

This paper describes experiments related to the application of well-known techniques of the texture feature extraction (Local Binary Patterns and Gabor filtering) to the problem of automatic face verification. Results of the tests show that simple image normalization strategy based on the eye center detection and a regular grid of fiducial points outperforms the more complicated approach, employing active models that are able to accurately locate several landmarks. On the other hand, the proposed shape descriptor provides promising results, while the texture features appear to be very sensitive to realistic illumination changes.

Authors

• dr inż. Maciej Smiatacz link open in new tab