Eye localization in low and standard definition content with application to face matching

In this paper we address the problem of eye localization for the purpose of face matching in low and standard definition image and video content. In addition to an explorative study that aimed at discovering the effect of eye localization accuracy on face matching performance, we also present a probabilistic eye localization method based on well-known multi-scale local binary patterns (LBPs). These patterns provide a simple but powerful spatial description of texture, and are robust to the noise typical to low and standard definition content. tensive evaluation involving multiple eye localizers and face matchers showed that the shape of the eye localizer error distribution has a big impact on face matching performance. Conditioned by the error distribution shape and the minimum required eye localization accuracy, eye localization can boost the performance of naive face matchers and allow for more efficient face matching without degrading its performance. The evaluation also showed that our proposed method has superior accuracy with respect to the state-of-the-art on eye localization, and that it fulfills the criteria for improving the face matching performance and efficiency mentioned above.