Design of N-Way DBN Classifier and Auto Encoder for Facial Similarity

Humera Tariq, Zainab Parveen


Recognizing a face is a routine task for human perception system, whereas building a similar computational model for Face recognition with 100% accuracy is quite challenging. This work is rich in terms of theory, literature, and experimentation with a novel approach towards face similarity technology. The paper concentrates on the design of an N-Way DBN Classifier and Similarity Metric DBN Auto-encoder for recognizing facial similarity (or Face Verification). PCA feature extraction process is combined with DBN to enhance the performance of the two proposed models. The models work under both Image Unrestricted and Image Restricted settings on the small LFW dataset. Many improvements and refinements needed to improve the models. With advancement in feature extraction stage, the proposed approaches might achieve competitive verification performance.


Face Recognition, Deep Belief Network, Restricted Boltzmann Machine (RBM), Auto Encoder

Full Text:


(C) 2010-2018 EduSoft