Person identification from heavily occluded face images

Abstract

In numerous occasions there is need to identify subjects shown in heavily occluded face images. Typical examples include the recognition of criminals whose facial images are captured by surveillance cameras. In such cases a significant part of the subjects face is occluded making the process of identification extremely difficult, both for automatic face recognition systems and human observers. In this paper we propose a face recognition algorithm, which can be used for identifying individuals with hidden facial parts. During the face recognition procedure, occluded facial regions are detected so that the model-based face recognition algorithm implemented makes use of information only from the non-occluded facial regions. With our approach information from occluded facial regions is not utilized during the process of face recognition hence the occlusions do not destruct the recognition process and as a result the probability of achieving correct identification is improved.