Towards a Robust Face Detector

In this work we present the preliminary results of a face detection system based on an hybrid approach: it combines typical feature-based techniques with image-based analysis, in order to better exploit the main characteristics available in the input image. Different modules contribute to the face detection task: 1) a template-based approach initially proposed in [12], 2) an edge-extraction technique well suited to deal with illumination-changes, 3) a multiple-classifier specifically designed to discard false positives and 4) a novel method based on a featureless representation of the eye-patterns that further improves the face/non-face discrimination. The experimental results show that the system can localize faces in images with complex background, even in presence of strong illumination changes.