IEEE/OSA/IAPR International Conference on Informatics, Electronics & Vision
An Efficient Face Detection in Color Images Using Eye Mouth Triangular Approach Md. Hafizur Rahman, Farjana Jhumur, Md. S. U. Yusuf, Tonmoy Das, Mohiuddin Ahmad Department of Electrical and Electronic Engineering Khulna University of Engineering & Technology Khulna-9203, Bangladesh Email:
[email protected],
[email protected] Abstract— Human face detection is important for the aspects of face recognition, real time eye tracking and video surveillance. In this paper, we develop face detection method in color images. These methods rely on a two step process. First, we detect human skin regions based on YCbCr color method with lighting compensation technique and nonlinear color transformation. To avoid the effect of brightness included in the RGB color space, we propose to model the skin color in the chromatic and pure color space YCbCr, which separates luminance and chrominance components. After detecting face region within the image we construct eye and mouth maps in order to work on eye / mouth geometry and orientation of eye / mouth triangle. The result works well on wide range of facial variation in color, position, scale and orientation with photo collection including both indoor and outdoor environments.
Keywords-face detection; color space transformation; detection; facial feature extraction ; eye mouth triangle
I.
skin
INTRODUCTION
Face is the most distinctive and widely used key to a person’s identity. In recent years, face detection has attracted much attention and its research has rapidly expanded by not only engineers but also neuroscientists, since it has many potential applications in computer vision communication and automatic access control system. Face detection and facial feature extraction have attracted considerable attention in the advancement of human-machine interaction as it provides a natural and efficient way to communicate between humans and machines. The problem of detecting the faces and facial parts in image sequences has become a popular area of research due to emerging applications in human-computer interface, surveillance system; secure access control, video conferencing, financial transaction, forensic applications, pedestrian detection, drivers alertness monitoring system, image database management system and so on. Numerous face detection techniques have been proposed to address the challenging issues associated with this problem in the literature. These techniques generally fall under four main categories of approach: knowledge-based, feature invariant, template matching, and appearance-based. Different techniques have been introduced recently, for example, pixel-based [1, 2], parts-based [3, 4], local edge
530
features [5, 6], Haar wavelets [3, 7] and Haar-like features [8, 9]. Earlier holistic representation schemes are able to detect faces [1,2], the recent systems with Haar-like features [8, 10] have demonstrated impressive results in detecting faces under occlusion. A large training set of face images is essential for the success of learning-based face detectors. The AdaBoost based face detector by Viola and Jones [8] demonstrated that faces can be fairly reliably detected in real time (i.e., more than 15 frames per second on 320 by 240 images with desktop computers) under partial occlusion. The weak classifiers are selected and weighted using the AdaBoost algorithm [11]. As there are large numbers of weak classifiers, they presented a method to rank these classifiers into several cascades using a set of optimization criteria. A method for detecting human faces in color images was presented in [12] that first determines the skin-color regions and then determined faces within those regions. A chroma chart was prepared to distinguish skin regions from non skin regions. In [13], a new approach was proposed for face detection based on skin color detection. It utilized the methodology of GMM to construct several skin color models for different kinds of skin colors. A robust and effective face identification system using trianglebased segmentation process was presented in [14] to extract face in various kinds of face images. Most of the aforementioned methods have some limitations dealing with human faces in these approaches. (1) They cannot detect a face which is smaller than 50×50 pixels. (2) They cannot detect multiple faces (more than 3 faces) in complex backgrounds. (3) They cannot handle the defocus and noise problems. (4) They cannot conquer the problem of partial occlusion of mouth or wearing sunglasses. Although there are some researches that can solve two or three problems as pointed out above, there is still no system that can solve all the mentioned problems. So, we propose a face detection algorithm which is able to handle a wide range of variations in static color images, based on a lighting compensation technique and a nonlinear color transformation. The main reason in comparison with the aforementioned methods for choosing this approach is that it is simple to implement, and the face detection rate is satisfactory. Another benefit is that time consumed for running this algorithm is relatively short, comparing with other methods such as neural network-based, SVM based face detector. The final reason for choosing this
ICIEV 2012
IEEE/OSA/IAPR International Conference on Informatics, Electronics & Vision algorithm is that there is no need to train the t system as face classifier. Face detectors based on neuraal network or its extension concepts are necessary to trainn the system for several hours, or even several days to obbtain an accurate result. Although it only need to train the systtem once to obtain the trained data, if the trained data lose or be damaged, this t carry out again. time-consuming training process must need to This proposed system detects faces usinng an eye mouth triangle. The proposed face detection system m can be the first step for any Automatic face recognitionn system, Human Computer Interaction systems, Video surveilllance systems etc. II.
METHODOLOGY Y
Our proposed face detection methods aree described in the flow chart given in figure 1. Input Color Image
Color Space Transformation
Skin Detection
Eye Detection Mouth Detection Eye/mouth Geometry and Orientation of Triangle
Output Figure 1. Overall proposed methood.
A. Color Space Transformation The approach on this paper will use mainnly the color based algorithm with the technique of color spaace transformation from RGB (red, green and blue) to YC CbCr (luminance, chrominance blue and red). The proposeed algorithm first locates the face region using skin-color. The T YCbCr color space is used to detect the skin region on thhe given input face image. The given input RGB image is converted into the YCbCr color space. Y = 0.299R + 0.587G + 0.114B Cb = -0.169R – 0.331G +0.500B + 128 Cr = 0.500R – 0.419G – 0.082B + 128 1
531
(1) (2) (3)
Figure 2.
Color space transsformation from RGB to YCbCr.
Color is a powerful cue of human h faces. The distribution of skin clusters is in a small regioon of the chromatic color space which is shown in Fig 3. Proccessing skin color is faster than processing other facial feaatures. Therefore, skin color detection is firstly performedd on the input color image to reduce the computational compplexity. The apparent difference in skin color perceived is mainly due to the darkness or fairness of the skin, characterized by the difference in the brightness of the color, which is governed by Y but not Cb and Cr in YCbbCr color space. Y, luminance component is brightness compponent, whereas Cb and Cr are chrominance components, which w correspond to color components. In the color deetection process, each pixel is classified as either skin or non-skin based on its color components. B. Skin Color Detection In the skin color detection prrocess, each pixel was classified as skin or non-skin based onn its color components. The detection window for skin coloor was determined based on the mean and standard deviationn of Cb and Cr component, obtained using 85 training faces in 10 input images. The Cb and Cr components of 85 facees are plotted in the color space in Fig.3; their histogram distribbution is shown in Fig. 4. • Establishing clear thrreshold in color space (explicit and direct model): As noted, skin color of individuals will fall in a small areea of color space. This threshold can be done very sim mply on a component or on a combination of severral components. Simplicity of this method is its advvantage and problems related to the human race, lightt and crowded backgrounds are its drawback and weakk point. 90