Fast and accurate localization of human faces

Abstract

Face detection is the first step in many face image processing applications. This work presents a simple, fast and robust method for reliable detection of faces in real-life images. We introduce a novel technique for distinguishing different skin colors, a capability that most of the other skin color based approaches lack. By distinguishing different shades of skin we are able to accurately detect facial regions even in complex scenes with near skin color backgrounds. Once the algorithm finds a candidate skin colored region, it verifies the region for being a face using a facial template composed of stable facial features. The algorithm can reliably detect multiple near-frontal, near-upright faces at different scales even in the presence of spectacles. We test the effectiveness of the method on various standard video sequences and two standard datasets for face detection.