Unconstrained and NIR Face Detection with a Robust and Unified Architecture

Abstract

This paper proposes a face detection method making use of Fast Successive Mean Quantization Transform (FSMQT) features for image representation to deal with illumination and sensor insensitive issues of the individual as well as the crowd face images. A split up Sparse Network of Winnows (SNoW) with Winnow updating rule is then exploited to speed up the original SNoW classifier. Features and classifiers are combined together with skin detection algorithm for fake face detection in crowd image and head orientation correction for near infrared faces. The experiment is performed with four databases, viz. BIOID, LFW, FDDB and IIT Delhi near infrared showing superior performance.