Lip-Based Identification Using YOLOR

In recent years, the exploitation of lip prints for biometric identification has gained much attention from the research community, with most of the efforts devoted to establishing that the physiological characteristic of the lip is discriminative and unique. Up until now research in this area has employed more classical feature engineering-based approaches and results that have been achieved are still not comparable with those yielded by more commonly used biometric characteristics. Furthermore, the field of lip detection is still an ongoing topic of research due to its many challenges which hinders the success of lip detection techniques. This work will determine the viability of newer methods on the task of lip detection and identification through the application of newer deep learning methods which is an apparent gap in this area. In this study YOLOR is applied on samples of faces from the CFD dataset to effectively achieve lip detection and identification. The results obtained are promising with a mAP of 99.5% and a precision and recall score of 67% and 99%, respectively.