Research On Face Detection And Facial Feature Location

Face detection and facial feature location are important branches in pattern recognition field. Face detection is a necessary first-step in face recognition systems, with the purpose of localizing and extracting the face region from the background. It also has several applications in areas such as content-based image retrieval, video coding, video conferencing, crowd surveillance, and intelligent human-computer interfaces. The human face is a dynamic object and has a high degree of variability in its appearance, which makes face detection a difficult problem in computer vision. Locating facial features is an important stage in many facial image interpretation tasks, such as face verification, face tracking or face expression recognition. In addition, the relevant theory and technology of face detection is not only the basis of face recognition and analysis, but also the key to resolve many questions of object detection and pattern recognition, such as vehicle detection, passerby detection et al. Therefore, the research of face detection and feature extraction has very important business and science value.This dissertation focuses on the research of face detection, facial feature location and their application to technique of driver active safe. In particular, the research focuses on the following topics:Two face detection methods based on the knowledge and support vector machine are proposed. The knowledge based methods encode human knowledge of what constitutes a typical face. Usually, the rules capture the relationships between facial features. This approach needn't training set like the approach based on machine learning. However, the problem of this approach is the difficulty in translating human knowledge into well-defined rules. If the rules are strict, they may fail to detect faces that do not pass all the rules. If the rules are too general, they may give many false positives. Therefore, combining the knowledge-based method with learning-based method can make use of their advantages. Face candidates can be extracted effectively by knowledge-based method, sequentially; the extracted regions are verified using a single SVM for making binary decision: face/non-face. This dissertation proposed two methods to extract face candidates. One is based on binary template matching; the other is based on rectangle feature. Experiments on BioID and MIT+CMU dataset show their effectivity.Face detection methods based on mixed features and Adaboost algorithm are presented. AdaBoost and Cascade algorithm are two poplar methods of training face detector. However, after the power of a strong classifier has reached a certain point, the non-face examples obtained by bootstrapping are very similar to the face patterns, in any space of the simple Haar-like features. It can be empirically shown that the classification error of Haar-like feature-based weak classifiers approaches 50%, and therefore boosting stops being effective in practice. To address this problem, a method combining local Haar-like features and global PCA features is proposed, and classifiers based on PCA features are ensembled to improve the performance. Based on this idea, a multi-view face detection algorithm is proposed. An improved detector structure is presented. The first classifier extracts faces of any pose from the background. Then more specific classifiers discriminate between different poses. Experiments have been done on MIT+CUM dataset.Facial feature location algorithms are proposed. Firstly, an eye variance filter is constructed to detect two eyes in the eye region which has been extracted in the eye pair location step. Then, another more robust algorithm is proposed, which extracts eyes by image entropy analysis and verifies the potential region by SVM. Mouth location is based on eyes location. Experiments have been done on BioID, JAFFE and ORL dataset.A computer-vision-based driver fatigue detection method is proposed. The method contains eye location, eye tracking, face tracking and eye state recognition. Eye location algorithm has been discussed. After eye detection, right eye is tracked based on particle filters. Recently particle filters algorithm has shown to be suitable to perform real-time tracking in cluttered environments. Unlike Kalman filters, which are limited to Gaussian probability distributions, particle filters are able to represent multimodal distributions. Face is also tracked by CamShift algorithm while eye is tracked, which is used as eye tracking verification. Eye state recognition is done in each frame, by which the decision whether the driver fatigue happens is done.