Multiple Face-Component Analysis: A Unified Approach towards Facial Recognition Tasks

Computer vision plays an important role in the AI dream of building a machine that can interact with humans as another person would. Among various computer vision tasks this machine would have to perform to interact effectively with humans, facial recognition tasks such as identity recognition, gender recognition, ethnicity recognition etc. are of immense importance. While several reasonably accurate algorithms for performing several of these individual tasks in isolation exist, not much work has been done in developing an algorithm for performing all of these tasks simultaneously. In this paper, we propose an algorithm that explores this idea. We have conducted experiments for the joint estimation of several components of the face such as identity, gender, ethnicity and expression on two well known datasets, namely, the Radboud faces dataset and the NimStim face stimulus dataset. The results we have obtained show the feasibility of considering facial recognition tasks under a single unified framework and that of building systems capable of multi-component recognition.