A STATE OF THE ART COMPARISON OF DATABASES FOR FACIAL OCCLUSION

Abstract

Face recognition continues to be one of the most popular research areas of image processing and computer vision. There are various face databases available to researchers for face detection and recognition. These databases are customized for a particular need of one algorithm. They are range in size, scope, and purpose. Few of these databases from the literature contain face occlusions in several positions of the faces to enable real world applications.  In this paper, we present four different occlusion face databases. These are Aleix-Robert (AR), Bosphorus, Labeled Faces in the Wild (LFW), and University of Milano Bicocca Database (UMB) face databases. At each section, the key features of the database are presented with the recording conditions, though not all of them are discussed at the same level of details. Detailed comparisons of the databases were made based on controlled and uncontrolled databases, 2D and 3D databases and also their uniqueness. Comparison was also made with other databases out of the categorization mentioned. The databases are useful for performing a rigorous benchmarking of face detection and recognition algorithms.