【作者】 高维飞；

【导师】 王春立；

【作者基本信息】 大连海事大学 ， 计算机科学与技术， 2011， 硕士

【摘要】 手语不仅包含手和手臂的动作,而且也包含表情、头部姿态和体势等动作。有实验表明,手语中只有手势而没有表情时,人所能理解的内容估计不超过60%。作为表情识别和头部姿态估计的前提,人脸检测具有重要的研究意义。本文针对手语视频中的人脸检测问题,结合Adaboost和肤色检测方法,改进手语视频中人脸检测的效果,并最终实现了一个高召回率和高准确率的人脸检测系统。具体的研究工作如下：1、本文首先研究了Adaboost人脸检测算法。针对手语视频中存在的手部干扰问题,本文在训练分类器时将大量人手样本加入到负样本集中,训练了一个正面人脸分类器。针对手语视频中头部呈现多姿态情况,本文又训练了侧面人脸分类器。实验结果表明Adaboost人脸检测算法具有较高准确率但召回率较低。2、本文研究了基于肤色的人脸检测算法。实验结果表明肤色人脸检测算法具有较高的召回率,但是其准确率和精度较低。3、本文将肤色人脸检测算法和Adaboost人脸检测算法相结合。利用Adaboost人脸分类器扫描肤色分割后的图像,精确定位人脸。实验表明,本方法不但有较高召回率,而且还有较高的准确率和精度。提高了手语视频的人脸检测效果。4、最后,本文设计与实现了结合Adaboost和肤色检测方法的人脸检测原型系统。更多还原

【Abstract】 Sign language contains not only hand and arm movements, but also other actions such as expression, head and body. The experiment shows that the contents of people understand are not more than 60% if the sign language has only gestures not expressions during the communication. As the premise of expression recognition and head pose estimation, face detection has the important research value. The solutions on the face detection problems are put forward in the sign language video in the paper, which combines the color of skin face detection algorithm with Adaboost face detection algorithm, and finally a face detection system is realized with a high recall rate and a high accuracy rate. Specific researches are as follows:1. Adaboost face detection algorithm is studied in this paper firstly. With the problems of the hand interference in sign language video, lots of hand samples have been put into the negative samples in training classifier, and a positive face classifier has been trained. Owing to the pose varied of head in sign language video, a profile face classifier is trained in this paper. Experimental results show that it has higher accuracy but lower recall rate.2. The color of skin face detection algorithm is studied in this paper. Experimental results show that color of skin face detection algorithm has higher recall rate, but lower precision and accuracy.3. To improve the face detection effect in the sign language video, the color of skin face detection algorithm and Adaboost face detection algorithm are combined. The skin-color segmentation image is scanned by Adaboost face classifier to locate face area accurately. Experimental results show that this method not only has high, and there’s a recall rate high precision and accuracy.4. At last, a face detection system, which combines the color of skin face detection algorithm with Adaboost face detection algorithm, has been designed.更多还原