Survey: How good are the current advances in image set based face identification? - Experiments on three popular benchmarks with a naïve approach

The recent proposed approaches on image set based face identification always follow a four-stage pipeline: face detection - face image representation - face image set modelling - identification; with face image set modelling being the additional step in this pipeline compared to that of the conventional image based face identification. As the research community moves forward, the performance in the area of image set based face identification have been slightly improved; however, the algorithms, mainly concentrated on the stages of face image set modelling and identification, have become dramatically complex. This paper shows that on the three most commonly used benchmarks, namely Honda/UCSD, CMU-MoBo and YouTube Celebrities datasets, a naïve Euclidean distance based approach can perform at least as good as, if not better than, the state-of-the-art algorithms. This leads to the question: how far has the current research tapped into the modelling of image face sets for the identification purpose?