Study On Techniques Of Face Detection And Facial Feature Location And System Development

In the dissertation, we carried on the study of face detection and facial feature location. As a key technique in the field of computer vision, face detection and facial feature location has many potential important applications in a wide range of fields including video surveillance, face recognition systems, content-based retrieval, advanced human and computer interaction. Learned from the domestic and international discourse and research papers concerning face detection and recognition in recent years, firstly we give a systematic summary of previous work. And we developed an automatic face detection system on the basis of the previous research on face detection. The system achieved crucial improvements on some traditional and classical algorithms of face detection and facial feature location. My primary work as following: 1. Based on the technology of real-time face detection method presented by Dr.Paul Viola, we designed and implemented the rapid frontal face detection algorithm with AdaBoost classifier and Haar-like features. An effective method of resizing the search window to enable us find the faces at different sizes was adopted which is more efficient than the traditional "pyramid"detection method. The algorithm achieves a high hit-rate higher than 99% on the standard face database AR and FERET. Especially, the detection speed of the system attains the criterion of real-time standard 15F/s. 2. In the paper, a computer vision color tracking algorithm is developed and applied to tracking human face. Firstly the algorithm creates a color model by taking 1D histogram from the H channel in HSV space. Then converting the incoming video pixels to a back project image, which is the probability distribution of flesh image. The color tracking algorithm shows a fast, efficient tracking quality combined with high performance for faces with variations in position, scale, pose and facial expression. 3. Using information fusion theory, we've completed the decision fusion of three classifiers: AdaBoost face detection classifier, color classifier and the contour character classifier. And we realized the feature fusion of different contour features using the linear weighted algorithm. Thus the information fusion scheme on face detection is finish. Experiments prove that this method can make the face detection more accurate and robust. 4. For static color image, we presented a facial feature location algorithm based on the YCbCr color space and the geometric relationship of the facial features. Firstly, we directly locate the eyes based on the eye map derived from eye chrominance map. Then using the color feature and the geometric relationship of facial features, we search and detect the mouth and nose location. Experiments show that this facial feature detection method is excellent both on detection speed and detection rate, which can attain the high detection rate higher than 95% both on IISL and IMMU face database.