A Comparative Study on Multi-view Discriminant Analysis and Source Domain Dictionary Based Face Recognition

摘要

Human face images captured in real world scenarios using surveillance cameras won't always contain single view, instead they usually contain multi-view. Recognizing multi-view faces is still a challenging task. Multi-view Discriminant Analysis (MDA) and Source Domain Dictionary (SSD) are two techniques which we have developed and analyzed in this paper to recognize faces across multi-view. In MDA the faces collected from various views are reflected to a discriminant general space by making use of transforms of those views. SSD on the other hand is based on sparse representation, which efficiently makes the dictionary model of source data. It also signifies each class of data discriminatively. Both the developed techniques are validated on CMU-Multi PIE face database which contains 337 people recorded under 15 different view positions and 19 different conditions.