El_Sayed M. Saad, M.M. Hadhoud, M.I. Moawad, M. El_Halawany, A.M. Abbas / Computing, 2007, Vol. 6, Issue 1, 25-34 25 A NEW SYMMETRY APPROACH FOR FRONTAL-VIEW FACE DETECTION El_Sayed M. Saad 1) , Mohiy M. Hadhoud 2) , Moawad I. Moawad 3) , Mohamed El_Halawany 3) , and Alaa M. Abbas 3)* 1) Faculty of Engineering, Helwan University, Egypt 2) Faculty of Computers and Information, Menoufia University, Egypt 3) Faculty of Electronic Engineering, Menouf, 32952, Menoufia University, Egypt * Corresponding Author: aladin_abbas@yahoo.com Abstract: An efficient algorithm for detecting frontal-view faces in color images is proposed. The proposed algorithm has a special task; it detects faces in the presence of skin-tone regions (human body, clothes, and background) Firstly, a pixel based color classifier is applied to segment the skin pixels from background. Next, a hybrid cluster algorithm is applied to partition the skin region. We introduce a new symmetry approach, which is the main distinguishing feature of the proposed algorithm. It measures a symmetrical value, searches for the real center of the region, and then removes the extra unsymmetrical skin pixels. The cost functions are adopted to locate the real two eyes of the candidate face region. A template matching process is preformed between an aligning frontal face model and the candidate face region as a verification step. Experimental results reveal that our algorithm can perform the detection of faces successfully under wide variations. Keywords: Face detection, image segmentation, clustering, cost functions, symmetry approach. 1. INTRODUCTION AND BACKGROUND Automatic face detection is an attractive research area. Its importance is due to its vital wide applications such as, criminal identification, security, and surveillance. Human face has lots of variations of image appearance, such as lighting conditions, different size, and simple/complex background either in still images or videos. Therefore, face detection is a great challenging task that should be overcome by engineers and scientists. Over the past years, the dominant mode of images was the gray mode, therefore many researchers proposed algorithms in simple background or complex background for gray images [1-10]. Recently, the advanced technology makes it easier to handle the color images by digital cameras, scanners, higher speed PCs with larger storage capacity, and broadband networks. As a result, color images have became the dominant mode and many researchers move their interest toward color images [11-13]. In general, the automatic face detection algorithms can be classified into two categories. The first category is based on the computation of geometric relationship among facial features. The second category is based on template matching. Human face detection algorithm has attracted the attention of many researchers. Jeng et al. [14] proposed an approach for detecting facial features. Firstly, the approach was enhancement the contrast by using a boost filter. A matching process started by randomly selecting two facial features as the eyes, and then a weighted evaluation function was computed, if the result was larger than a certain threshold, these two features accepted as eyes. The drawbacks of this approach are the limitations of face size; not smaller than 80x80, and the image must contain only one face. Lin and Fan [15] presented a triangle-based approach for the detection of human faces. They extracted the potential face regions from the input image, thereafter any 3 centers of different blocks form an isosceles triangle were detected. Weighting mask function was applied to decide whether a potential face region contains a face. The algorithm failed to deal with too dark images or occluded eyes. Wu and Zhou [16] introduced a face selector method. The eye was segmented by finding regions that are roughly as large as real eyes and darker than their neighborhoods. Some cases in which the face selector failed: one eye was near the image border, presence of glasses, dark images, and high rotation angle. Shih and Chuang [17] proposed an approach for extracting human head by high threshold image, computing@tanet.edu.te.ua www.computingonline.net ISSN 1727-6209 International Journal of Computing