Likelihood Ratio-Based Detection of Facial Features
Asker M. Bazen, Raymond N. J. Veldhuis, Gerrie H. Croonen · University of Twente Research Information · 2003
One of the first steps in face recognition, after image acquisition, is registration. A simple but effective technique of registration is to align facial features, such as eyes, nose and mouth, as well as possible to a standard face. This requires an accurate automatic estimate of the locations of those features. This contribution proposes a method for estimating the locations of facial features based on likelihood ratio-based detection. A post-processing step that evaluates the topology of the facial features is added to reduce the number of false detections. Although the individual detectors only have a reasonable performance (equal error rates range from 3.3% for the eyes to 1.0% for the nose), the positions of the facial features are estimated correctly in 95% of the face images.