6533b861fe1ef96bd12c45da

RESEARCH PRODUCT

Real Time Robust Embedded Face Detection Using High Level Description

subject

description

Face detection is a fundamental prerequisite step in the process of face recognition. It consists of automatically finding all the faces in an image despite the considerable variations of lighting, background, appearance of people, position/orientation of faces, and their sizes. This type of object detection has the distinction of having a very large intra-class, making it a particularly difficult problem to solve, especially when one wishes to achieve real time processing. A human being has a great ability to analyze images. He can extract the information about it and focus only on areas of interest (the phenomenon of attention). Thereafter he can detect faces in an extremely reliable way. Indeed, a human being is able to easily locate faces in its environment despite difficult conditions such as occlusions of parts of a face and bad lightening. Many studies have been conducted to try to replicate this process, automatically using machines, because face detection is considered as a prerequisite for many computer vision application areas such as security, surveillance, and content based image retrieval. Over the last two decades multiple robust algorithmic solutions were proposed. However, researches in the field of computer vision and pattern recognition in particular tend to focus on the algorithmic and functional parts. This generally leads to implementations with little constraints of time, computing power and memory. Most of these techniques, even if they achieve good performance in terms of detection, are not suited for real time application systems. Nonetheless, Boosting–based methods, firstly introduced by Viola and Jones in (Viola & Jones, 2001; 2002), has led the state-of-the-art in face detection systems. These methods present the first near real time robust solution and by far the best speed / detection compromise in the state-of-the-art (up to 15 frames/s and 90% detection on 320x240 images). This family of detectors relies upon a cascade of several classification stages of progressive complexity (around 20-40 stages for face detection). Depending on its complexity, each stage contains several classifiers trained by a boosting algorithm (Freund & Schapire, 1995; Lienhart, Kuranov, & Pisarevsky, 2003; Viola & Jones, 2002) These algorithms help achieving a linear combination of weak classifiers (often a single threshold), capable of real time face detection with high detection rates. Such a technique can be divided into two phases: Training and detection (through the cascade). While the training phase can be done offline and might take several days of processing, the final cascade detector should enable real-time processing. The goal is to run through a given