A Comparative Analysis for the Performance of LFW and ORL Databases in Facial Recognition

It has become clear in the last many years that face recognition has attracted the attention of researchers. Until now, this has been a difficult area of research due to several problems. The face continues to be the most difficult subject of study for experts within the realm of computer vision and image processing, because it is an element with diverse sensory properties. There are several face images databases used to train and test facial recognition systems used in the recent literature with different results. But until now, the most effective database has not been identified. The free facial image databases LFW (Labeled Faces in the Wild) and ORL (Olivetti Research Laboratory) are the most widely used. The main idea of this survey work is to compare the selected face databases based on using different face recognition methods (different feature extraction and classification techniques).The performance, accuracy, and computational requirements of these methods are analyzed through a series of case studies and empirical evaluations with a confusion matrix. The results of this comparison will be used in the author's future work in the face detection and recognition software environment for evaluating performance and testing purposes. The results obtained support that the suitable choice between LFW and ORL databases depends on the specific goals of the facial recognition system. For initial development and controlled testing, ORL is highly effective. For evaluating performance in real-world scenarios, LFW is more representative but challenging.