Research Of Face Detection And Eye Location Algorithm

As a key technology in face information processing, face detection attracts a widespread attention in the field of computer vision and pattern recognition in recent years because of its high value of academic research and commercial applications. With the development of intelligent information processing technology, face detection can be applied to the identification, video encoding, content-based retrieval, automatic control, human-computer interaction and so on.Face detection and location is the indispensable part of Face Recognition System and the premise of the face recognition technology. The existing typical methods of face detection are summarized and analyzed. Furthermore Contourlet transformation which has a strong ability of extracting marginal features is applied to the extraction of image features firstly. And a new method of representing face images sparsely, so-called face pyramid, is presented in this paper. The method of face pyramid makes full use of the information of high and low frequency and direction which is produced using the decomposition based on Contourlet transformation, and extracts face features of great identification using a small quantity of coefficients. And then based on the Adaboost face detection algorithm the training and detecting process is optimized respectively. Finally a cascade classifier is trained using the developed algorithm; thereby realizing the detection of frontal face from complex face image. Compared with Adaboost face detection algorithm the number of the face-pyramid based feature is much less than the number of Haar features. So the cascade classifier in this paper decreases the training time greatly besides guaranteeing a high detection precision.The two eye centers are located correctly using integral projection algorithm and then the face images are normalized to prepare for the face recognition. In order to locate the eye centers more precisely some image pre-processing technologies are used to protect the integral projection curve from noise; thereby locating the eyes precisely.The experiments demonstrate that the cascade classifier algorithm can detect the frontal face and some rotated face on a complex background efficiently. This algorithm is proved to be robust and suitable for the change of face expressions, the light condition of picture and accessories (glasses, hats) etc. The eye location algorithm can locate the eye centers in an image of frontal face precisely.The face detection module is constructed by combining the face detection and location as the pre-order part of the face recognition system. Based on the module a face recognition system is developed as well. Our program about face detection and location has a good portability and lays a good foundation for future research and application.