Rule-Based Face Detection in Color Images using Normalized RGB Color Space – A Comparative Study Ravi Subban Department of Computer Science School of Engineering and Technology Pondicherry University Puducherry, INDIA sravicite@gmail.com Richa Mishra Department of Computer Science School of Engineering and Technology Pondicherry University Puducherry - 605014, INDIA richamishra.0312@gmail.com Abstract—Detecting human faces in color images plays an important role in real life application such as face recognition, human computer interface, video surveillance and face image database management. This paper provides a comparative study on some of the human skin segmentation techniques. The linear piecewise decision boundary strategy is used for human skin detection techniques with the help of normalized RGB color space. Both adaptive and nonadaptive skin color models are used for skin detection. The rule based technique using the normalized RGB is used to detect facial features like lips and eye. This technique is very effective in detecting facial features. The experimental results show that nonadaptive skin color models produces better results as compared to the adaptive skin color models. Keywords—Face Detection, Skin Detection, Normalized RGB, Facial Features, Adaptive Approach, Nonadaptive Approach. I. INTRODUCTION Human face detection in color images is an important research area in the fields of pattern recognition and computer vision. It is a preprocessing step in the fields of automatic face recognition, video conference, intelligent video surveillance, advance human-computer interaction, medical diagnosis, criminal and terrorist identification [5]. It remains elusive because of variations in illumination, pose, expression and visibility, which complicates the face detection process, especially under real-time constraints [6]. Face detection techniques are used to determine the presence and location of a face in an image, by distinguishing it from the background of the image, or from all other patterns present in the image. Face localization can either be a preprocessing step or an integral part of face detection activity. A number of approaches are in use for detecting facial regions in color images [16]. One of the important approaches used to detect face regions in color images is through skin detection, because skin-tone color is one of the most important features of human faces. The skin regions in a color image can be segmented by processing one pixel at a time sequentially and independently. Skin segmentation means differentiating skin regions from non-skin regions in a color image [25]. Although there are many color spaces that can be used for skin detection, the normalized RGB has been used in this paper because this color space is more suitable for skin detection and facial feature detection. It is also used in this paper due to the evidences that human skin color is more compactly represented in chromaticity space than in other color spaces [2]. Face detection techniques can be classified into two main categories: feature-based and image-based. The complete information of face is one of the necessary requirements of face detection technique. Feature-based techniques depend on feature derivation and analysis to gain the required knowledge about faces [4]. Face detection is inadequate without detecting its facial features. It generally includes salient points, which can be traced easily, like corner of eyes, nostrils, lip corners, ear corners, etc. The facial geometry is generally estimated based on the position of eyes. On the other hand, image-based techniques treat face detection as a general pattern recognition problem. It uses training algorithms to classify regions into face or non-face classes [4]. Jones et al. [6] construct a generic color model as well as separate skin and non-skin models using RGB color space over large image datasets that are available on the World Wide Web. Jayaram et al. [3] also has used large dataset for the making the comparison for combinations of nine color spaces in the the presence or the absence of the illuminance component, and the two coloring models. The methods described in this paper use feature-based technique to identify the human face regions in the color images. This paper considers all the methods used for the face detection based on feature-based technique in terms of their effect on the color images of the database that is commonly used. The remainder of the paper is organized as follows: Section II describes the previous work done on human skin color detection and face detection using normalized RGB. Section III focuses on human skin detection techniques. The section IV focuses on experimental results and describes about the effect of methods on the color images. Finally, section V concludes the paper. II. RELATED WORK Brand et al. [7] assessed the merits of three different approaches to pixel-level human skin detection over RGB color space. Pankaj et al. [10] has presented a fundamental unbiased study of five different color spaces in addition to normalized RGB, for detecting foreground objects and their 978-1-4673-1344-5/12/$31.00 ©2012 IEEE