Face Detection And Location Method

Recognition and analysis of human faces can be widely used in such fields as personal identification recognition, safety inspection, human-computer interaction, expression analysis and lip-reading. As the basis, necessity and prerequisite of human face recognition and analysis, human face detection with the help of computer attracted people's attention at a very early time. Along with the popularization of computer application, the improvement of computer performance and achievements in image procession and pattern recognition, human-face related applications are getting nearer and nearer to practical use, thus resulting in more and more emphasis over research for human face detection.Facial features detection and localization, an important technique in human face analysis, is specialized in searching for facial features (eyes, nose, mouth, ears, etc) with in a given region in an image or image sequence. It finds applications in various areas, such as face detection, face recognition, gesture recognition, expression recognition, face image compression and reconstruction, and face cartoon. This thesis attempts to give an overview of the latest development in this field by classifying the newly proposed methods into five categories, namely the ones based on knowledge, geometry information, color, appearance and relative location. All these methods appeared in the related papers or works published on international, as well as Chinese, journals and conference proceedings in recent years. Then their performances in accuracy, robustness and computational expense are roughly estimated and compared, and some discussions about the criteria on which the estimation and comparison are based provided.There have already been a number of successful algorithms for human-face automatic detection and localization. However, There still exist some problems to solve about face localization in complex poses. The thesis proposes two algorithms for automatic face detection and localization. Algorithm I is based thefeatures of eyes. And algorithm II is based the symmetry of human faces.In algorithm I features of eyes, are integrated with a skin color model, morphological processing, eye detection and localization. First, skin regions are detected based on a model of skin color; Second, they are further segmented into a series of skin blocks. Each block is morphologically processed in order to smooth out the biconvex and small backdate and to fill holes. At the same time, connectivity of different blocks can be avoided in this way; Finally, the eyes are located if a block is a face block using the features of human eyes. Algorithm I can be used to detect multiple faces with different poses in a color image. It uses complexion model and the features of eyes to preliminarily extract human-face regions, and shows higher speed for detection than the existing algorithm.Algorithm II is proposed for face localization in complex poses. The main idea is to use the symmetry of human faces. First, skin regions are detected based on a model of skin color; Second, Based on the observation that eye area is of low gray level and great gradient, some candidate regions of eyes can be obtained and then matches of these regions are performed to find the best ones. The algorithm can be used to detect multiple faces with different poses or expressions, even with glasses in a color image.The thesis respectively gives the experimental results of algorithm I and algorithm II ,and compares the detection and localization performance of the two algorithms with existing approaches. Experimental results show that the proposed algorithms are superior over existing ones in the robustness and effectiveness of the method in complex condition.