International Journal of Advanced Computer Science, Vol. 3, No. 8, Pp. 428-433, Aug., 2013. Manuscript Received: 10,May, 2013 Revised: 25,May, 2013 Accepted: 9, Jul., 2013 Published: 15, Jul., 2013 Keywords Gender detection, face detection, object detection, image processing Abstract Computer vision-based gender detection from facial images is a challenging and important task for computer vision-based researchers. The automatic gender detection from face images has potential applications in visual surveillance and human-computer interaction systems (HCI). Human faces provide important visual information for gender perception. This research presents a novel approach for gender detection from facial images. The system can automatically detect face from input images and the detected facial area is taken as region of interest (ROI). Discrete Cosine Transformation (DCT) of that ROI plays an important role in gender detection. A gender knowledgebase of the processed DCT is created utilizing supervised learning. To detect gender, input image is passed through a classifier which is based on that knowledgebase. To improve the matching accuracy Local Binary Pattern (LBP) of the ROI is done before converting it into DCT. This research has experimented on a database of more than 4000 facial images which are mainly of this subcontinent in order to evaluate the performance of the proposed system. The average accuracy rate achieved by the system is more than 78%. 1 1. Introduction Recently gender classification from face image is an attractive research topic and one of the actual problems of computer vision. If a person prevent from being seen his hairstyle, remove his facial hair and makes other changes to his face, human can still detect the sex with an accuracy of more than 90%. This observation attracts most of the scientist that what is the facial information by which human can make difference between men and women. A human can detect gender from face easily but it is very challenging task for computer which has no intelligence. In modern world everything is going to be machine dependent. With the growing demand for security, reliability, convenience, computer vision approaches such as face detection, gesture detection, person identification, motion detection and perhaps most fundamentally gender detection will play important role in our life. Target of this research is to develop This work is supported by University of Dhaka, Bangladesh. 1,  Emon Kumar Dey is with ...(Email: emonkd@unívdhaka.edu), 2,  Mohsin Khan is with ...(Email: mohsíncsedu@yahoo.com), 3,  & Md Haider Ali is with ...(Email: haíder@unívdhaka.edu). an intelligent system for identifying the gender of a person by the facial image. Sectors such as biometric authentication, high-tech surveillance, security systems, criminology, automatic psycho physiologic inspection, etc. can be benefited by the automatic gender detection system. There are mainly two approaches. The first one came from the psychophysical explorations of human face. It generally used the features such as distances between nose, eyes and mouth, areas of different face parts and so on. The second approach use the low level information of the face image which is based on the image pixels of the face. The first approach also known as the high level features based classification. But this first method has some problem for automatic detection. The result is not very accurate and unsolved in many cases. This research is mainly focused on the second approach which is based on low level information approach. This approach examines the pixel properties of different area of a face image and makes decision about the gender of the face image. 2. Related Works The sex of a face is conveyed by several classes of information, namely (a) local features (such as facial hair, eyebrows, and skin texture), (b) configural relationships between features, and (c) the 3D structure of the face. The shape differences between the two sexes, and compared the average male and average female heads which were obtained using a laser range scanner. On average, the male face has a more protuberant nose, brow, chin/jaw than the female face. The female face, on the other hand, has somewhat more protrusive cheeks than the male face. Moreover, the greatest differences were found in the regions of the nose and chin. Cottrell et al [1] proposed a method which reduced the dimension of whole face images by applying auto encoder network and classified gender based on the reduced input features. Tamura et al [2] used a neural network and proved that if the image is even very low resolution such as 8 x 8 can be used for gender classification. Jain and Huang [3] extracted the features by an approach which known as independent component analysis (ICA) and classified it with linear discriminate analysis (LDA). Burton et al.[4] extracted point-to-point distances from 73 fixed points on face images and used discriminant analysis as a classifier. Brunelli and Poggio (1992) extracted 16 geometric features such as eyebrow thickness and pupilto- eyebrow distance and used HyperBF networks as a classifier. BenAbdelkader and Griffin [5] proposed a method. This method extracted regions from the face. The region is then used for the input of Computer Vision-Based Gender Detection from Facial Image Emon Kumar Dey, Mohsin Khan, & Md Haider Ali