复杂环境下的人脸识别增强算法

Face Recognition Algorithm in Complex Environment of Mobile Platform

摘要: 本文针对复杂环境下人脸识别速度慢准确性低的问题进行研究，提出了一种人脸识别算法。首先采用Adaboost与肤色分割相结合的方式进行人脸检测。利用肤色分割技术速度快的优点筛选出候选人脸区域，同时采用Haar特征来训练分类器，进一步加强检测性能，提升复杂背景下的准确性。之后采用基于线性判别式分析的人脸识别方法(LDA)，LDA算法是对样本进行分类的监督学习方法，通过对样本的标签信息进行运用，寻找最佳投影方向子空间来完成分类。实验结果表明，本算法在复杂背景下识别速度与准确性上都有很大的提升。

Abstract: This paper studies the problem of slow and low accuracy of face recognition in complex environment and proposes a face recognition algorithm. Firstly, Adaboost combined with skin color segmentation for face detection. The advantages of fast skin color segmentation technique are used to screen out candidate face areas. Meanwhile, Haar features are used to train classifiers to further enhance detection performance and improve accuracy in complex background. Then, the face recognition method (LDA) based on linear discriminant analysis is adopted. LDA algorithm is a supervised learning method to classify samples. By applying the label information of samples, the optimal projection subspace is found to complete the classification. Experimental results show that the algorithm improves the recognition speed and accuracy greatly.