Deep Learning Model for Face Recognition in Unconstrained Environment
Face recognition from videos is challenging problem as the face image captured has variations in terms of pose, Occlusion, blur and resolution. It has many applications including security monitoring and authentication. A subset of Indian Movies Face database (IMFDB) which has collection of face images retrieved from movie/video of actors which vary in terms of blur, pose, noise and illumination is used in our work. Our work focuses on the use of pre-trained deep learning models and applies transfer learning to the features extracted from the CNN layers. Later we compare it Fine tuned model. The results show that the accuracy is 99.89 using CNN as feature extractor and 96.3 when we fine tune the VGG-Face. The Fine tuned network of VGG-Face learnt more generic features when compared with its counterpart transfer learning. When applied on VGG16 transfer learning achieved 93.9.