COMPACT: biometric dataset of face images acquired in uncontrolled indoor environment

Abstract

Biometric databases are important components that help to improve state-of-the-art recognition performance. The availability of more and more difficult data attracts the researchers' attention, who systematically develop novel recognition algorithms and increase identification accuracy. Surprisingly, most of the popular face datasets, like LFW or IJBA are not fully unconstrained. The majority of the available images were not acquired on-the-move, which reduces the amount of blur caused by motion or incorrect focusing. Therefore, in this paper, the COMPACT database for studying less-cooperative face recognition is introduced. The dataset consists of high-resolution images of 108 subjects acquired in a fully automated manner as people go through the recognition gate. This ensures that the collected data contains the real world degradation factors: different distances, expressions, occlusions, pose variations and motion blur. Additionally, the authors conducted a series of experiments that verify face recognition performance on the collected data.