Fast and Accurate Face Recognition With Image Sets

Hakan Cevikalp, Hasan Serhan Yavuz; Proceedings of the IEEE International Conference on Computer Vision (ICCV), 2017, pp. 1564-1572

Abstract


In this study, we propose a fast and accurate method to approximate the distances from gallery images to the region spanned by the query set for large-scale face recognition applications using image sets. To this end, we introduce a new polyhedral conic classifier that will enable us to compute those distances efficiently by using simple dot products. We also derive one-class formulation of the proposed classifier that can use query set examples only. This makes the method ideal for real-time applications since testing time approximately becomes the independent of the size of the gallery set. One-class formulation can also be used in a cascade system with more complex and time-consuming methods to return the most promising candidate gallery sets in the first stage of the cascade so that more complex methods can be run on those a few candidate sets. The proposed methods achieve the best accuracies on all tested small and moderate sized datasets.

Related Material


[pdf]
[bibtex]
@InProceedings{Cevikalp_2017_ICCV,
author = {Cevikalp, Hakan and Serhan Yavuz, Hasan},
title = {Fast and Accurate Face Recognition With Image Sets},
booktitle = {Proceedings of the IEEE International Conference on Computer Vision (ICCV) Workshops},
month = {Oct},
year = {2017}
}