Research Of Multi-view Face Detection

Face detection which is the basis of face recognition technology, has become the hot spot of research and intelligent application in the field of computer vision and pattern recognition. Since the pioneering work of Viola and Jone, the technology of frontal face detection has reached the level of practical application, but for multi-view face detection, there is still not any solution meet the needs of practical application, the latter research lag becomes an obstacle which impede the face recognition technology being reached truly practical application level.This paper focused on multi-view face detection methods, including a variety of algorithms research, sample database establishment and eventually achieve a multi-view face detection system in a controlled illumination conditions and complex background environment. The specific research contents include the following aspects:1. The multi-view face images and non-face images has been collected as the multi-view face sample database, and the several adaboost classifier which can effectively classify a particular multi-view face has been trained on this face database.2. Based on the frontal face detection method proposed by Viola and Jone, some improvements has been proposed. The face database and the test images has been converted to the texture images using LBP method which can effectively use facial texture features that distinguish of face and non-face region, but also have good robustness on light that can improve the stability of the system.3. The method of multi-view face detection based on skin color segmentation and geometric features combination with texture features has been proposed. The pose of face does not effect the extraction of skin color region. Face region can be extracted by the skin color model, then the geometric features and texture features of this region can be extracted using Hu moments and LBP method. Finally, multi-view face classifier which can effectively detect different poses faces can be trained by the support vector machine.4. the final multi-view face detection system can be realized by combining the above mentioned methods based on color and adaboost, so as to further improve the accuracy of face detection.