The Facial Recognition System has come a long way. Its usage is crucial in quite a few applications, for instance - photo retrieval, surveillance, authentication/access control systems etc. But, it is yet to completely overcome the challenges which have constantly played with its quality of delivery. Listed below are the challenges which limit the potential of a Facial Recognition System to go that extra mile.
IlluminationFor instance, a slight change in lighting conditions has always been known to cause a major impact on its results. If the illumination tends to vary, then; even if the same individual gets captured with the same sensor and with an almost identical facial expression and pose, the results that emerge may appear quite different.
BackgroundThe placement of the subject also serves as a significant contributor to the limitations. A facial recognition system might not produce the same results outdoors compared to what it produces indoors because the factors - impacting its performance - change as soon as the locations change. Additional factors, such as individual expressions, aging etc. contribute significantly to these variations.
PoseFacial Recognition Systems are highly sensitive to pose variations. The movements of head or differing POV of a camera can invariably cause changes in face appearance and generate intra‐class variations making automated face recognition across pose a tough nut to crack.
OcclusionOcclusions of the face such as beard, moustache, accessories (goggles, caps, mask etc.) also meddle with the evaluation of a face recognition system. Presence of such components make the subject diverse and hence it becomes difficult for the system to operate in a non-simulated environment.
ExpressionsAnother significant factor which needs to be taken into account is different expressions of the same individual. Macro and micro expressions find their place on someone's face due to changes in one's emotional state and in the wake of such expressions - which are many - the efficient recognition becomes difficult.
ComplexityExisting state-of- the-art methods of facial recognition rely on ‘too-deep’ Convolutional Neural Network (CNN) architecture which are very complex and unsuitable for real-time performance on embedded devices.
An ideal Face recognition system should be tolerant to variations in illumination, expression, pose and occlusion. It should be scalable to large number of users with need for capturing minimal images during registration while doing away with complex architecture at the same time.
Join us @Embedded Vision Summit to discover how a low-complex, embedded friendly CNN architecture can be used for face recognition. A Face Recognition System that’s independent of image variations and real-time on embedded devices