基于Adaboost算法的人脸检测系统设计与实现【毕业论文】.docVIP

本科毕业论文 （20 届） 基于Adaboost算法的人脸检测系统设计与实现 专业：计算机科学与技术 摘 要 Adaboost是一种迭代算法，这一算法的核心思想是针对同一个训练集训练不同的分类器(弱分类器)，然后把这些类器集合起来，构成一个更强的最终分类器(强分类器)。Adaboost算法是Freund和Schapire根据在线分配算法的基础上提出的，他们详细分析了Adaboost算法错误率的上界，以及为了使强分类器 达到错误率，算法所需要的最多迭代次数等相关问题。使用Adaboost分类器能够排除一些不必要的训练数据特征，并将主要注意力放在关键的训练数据上面。 在现在的社会中有很多的方法来核实一个人的身份，人脸识别是当前人工智能研究的热点，到目前为止有很多相关成果。而要想进行人脸识别，首先就要对人脸进行检测。对于静态图像中人脸检测这一问题，可以结合Adaboost算法来进行。因为现在已经有较成熟的理论和研究成果，所以Adaboost算法可以很好的解决人脸检测问题。 本文利用Harr特征、积分图以及Adaboost算法来解决静态人脸检测问题。通过Harr特征原型，利用积分图提取出特征值，在用Adaboost迭代算法对训练样本进行计算得到所需参数对待测图像进行检测人脸，如果检测到人脸则返回人脸在图像中的位置。实验表明，Adaboost算法能有效地解决人脸检测这类问题。 【关键词】Adaboost算法，Harr，积分图，训练样本，人脸检测 Design and Implementation of Face Detection System Based on Adaboost Algorithm Abstract As an iterative algorithm, the core idea of Adaboost algorithm can train various classifiers (weak classifier) aiming at the same training set, then these classifiers were assembled together to form a stronger final classifier (strong classifier). Adaboost algorithm is originated by Freund and Schapire based on online distribution algorithm based on a detailed analysis of their error rate on the Adaboost algorithm community, In order to achieve low classification error rate, the algorithm needs a upper boundary of iteration number and other related issues. Adaboost classifier can use to remove some unnecessary features of training data and training the main focus on the critical data on it. In society, there are many ways to verify a persons identity, face recognition is currently a hot research in artificial intelligence, so far there are many relevant results. However, if we want to do the face recognition, first,we must detect the human face. For static images of human face detection, we can combined Adaboost algorithm to do the face recognition. There are more mature theory and research, so Adaboost algorithm can solve the problem of face detection. In this paper, we use Harr features, integration plans and Adaboost algorithm to solve the static face detec