Locating human faces within images

Abstract

This paper presents an intelligent system to locate human faces within images. The proposed system can handle facial pattern variations due to certain changes in pose, illumination, and expression, as well as existence of spectacles, facial-hair, and occlusion. The system consists of three modules: preprocessing, face-components extraction, and final decision-making. In the first module, image processing algorithms are performed on images captured by cameras. Face components are extracted in the second module. A fuzzy neural network-based algorithm is designed for this purpose. In the last module, a commonsense-knowledge base is used for final evaluation of the identified features and determination of the face locations. The performance of the system is evaluated by conducting experiments on seven large test sets.

Introduction

Locating human faces in images generated by cameras is a complicated task to automate [21]. The reason for this is that there are several variables involved which can affect image of a face. The hindrance to face detection is the presence of additional features such as eyeglasses, moustache, or beard. The presence of these features can greatly distort the basic image of a face. In addition, the surrounding environment can also pose a problem. Changes in light-source distribution can cast or remove significant shadows from a particular face, therefore bringing more variability to facial patterns. A face locator should cope with the wide range of allowable facial pattern variations in an image.

In this paper, an intelligent face locator is proposed to be used for facial tracking in various applications including humanoid robots. The proposed face locator can deal with different variations in an image. The face locator improves the performances of the existing systems by accommodating for several variations in the input image such as illumination, expression, pose, facial-hair, occlusion, noise, eyeglasses, and background. Some of these image variations, e.g. illumination, are handled through including examples of the image variations in the training sets. Other image variations are handled through specifically design of the face locator.

The face locator consists of three modules: preprocessing, face-components extraction, and final decision-making. In the first module, standard image preprocessing is performed to enhance and restore the raw input image. Face components are then extracted in the second module using fuzzy neural networks. In the third module, final evaluation of the identified face components and determination of the face locations are performed using a commonsense-knowledge base.

This paper is organised as follows. The existing face-detection methods are reviewed in Section 2. In Section 3, the architecture of the proposed system is given in detail. The experimental results on a large collection of several test images are presented in Section 4. In Section 5, the performance of the face locator is compared against those of two well-known existing face-detection systems. In Section 6, concluding remarks are given.