Normalization method for misaligned face samples in non-cooperative biometrics

Reliability and robustness of a verification system is a major concern in biometrics. One of the most important covariates in face non-cooperative biometrics is misalignment, which exist because of errors in face detector. Current research is happening in this direction in order to make non-cooperative biometrics more reliable and consistent. In this paper, we have proposed the face normalization method to overcome the negative effects of misalignment. This method is based on first determining the eye distance and then determining the face which depend on the location of eyes. With this method, the recognition accuracy increases significantly from 44.95% to 66.66 %. It is worth mentioning that the experiments were conducted on the dataset, where large variations of poses, scales and illumination exists and these characteristics makes this dataset more appropriate for doing validation for non-cooperative face biometrics. However, detection rate with eye detector will be smaller than that of face detector. Thus, it's essential to make eye detector more reliable and robust in non-cooperative face biometrics.