University of Macau Abstract Face Detection in Skin Color Modeling and Template Matching by Wong lok Lan Thesis Supervisor: Professor Wu Enhua E-Commerce Technology Face Detection is originated from face recognition, as required from the development of various applications such as security and surveillance system, human-computer interaction and computer vision system. Face recognition becomes an essential topic in the technical development and research. Moreover, it acts as a vital role to build up a skin-color modeling system. In this thesis, we will utilize two methods to build up skin-color modeling: (1) Color Space and (2) Gaussian Mixture Model. Besides, we will also make the explanation on the extraction of skin regions, the data elements for the interface, the design of the user interface and the optimization of MATLAB program and so on. Among Chapters, firstly, we will use examples and figures to explain and present the contents and basic theory of the items mentioned above. Secondly, we will simulate how to realize the skin-color modeling and the extraction of skin regions by MATLAB. Finally, we will prove that (1) skin-color model based on Color Space is an efficient method to deal with the still color image under certain restrictions. (2) skin-color model based on Gaussian Mixture Model also can deal with the restricted photos. In addition, it also can be used to detect human faces in photos which contain multi faces which are taken within illumination or other extreme conditions. In addition, we will also discuss the reliability of the method: template matching in this thesis. From our system, we will focus on non-skin color region filtering. By using the method of skin color segmentation to reject the non-skin color regions. Then, by applying morphological processing to end up with candidate images, which only contain skin regions. Then, the system is devoted to the detection of human face. After performing the process in the first part, we get the candidate images. However, these images still contain some non-face regions (such as: arms, legs, hands, etc. and the background image which the color is similar to human skin). Therefore, in the second part, we need a threshold to distinguish the human face regions which are ticked out with the size, the proportion of width and length of region, the proportion of skin pixels and Euler number. Then, by using a human face template to detect the human face and to locate the coordinates (× and y axis) in the image.