基于SURF与Hough森林的人脸检测研究  

Face Detection Based on SURF and Hough Forests

作　　者：严明君[1,2] 项俊[3] 罗艳[1,2] 侯建华[1,2] 

机构地区：[1]中南民族大学智能无线通信湖北省重点实验室,武汉430074 [2]中南民族大学电子信息工程学院,武汉430074 [3]华中科技大学自动化学院,武汉430074

出　　处：《计算机科学》2014年第7期301-305,F0003,共6页Computer Science

基　　金：国家自然科学基金项目(61141010);武汉市科技供需对接计划项目(201051824575);湖北省自然科学基金项目(2012FFA113);中南民族大学中央高校基本科研业务费专项资金项目(CZY13033)资助

摘　　要：为实现复杂场景中的人脸检测与定位,提出了一种基于快速鲁棒特征SURF与Hough森林的人脸检测算法。采用SURF局部特征构建Hough森林分类器,每个叶子节点存储类别信息与特征点到达目标中心的偏移量,在图像局部外观与Hough投票之间建立映射关系,生成有监督的判别式的码本,获得可靠的概率Hough投票,以此预测目标中心位置,提高了检测精度。与此同时,采用SURF局部特征提取图像兴趣点有助于减小计算量、加快检测速度。实验证明了所提算法的有效性。In order to realize the face detection and localization in complicated scenes, this paper presented an algorithm for face detection based on SURF （speeded up robust features） and Hough forests. SURF local features were adopted to construct a Hough forest classifier. Each leaf node stored the class information as well as the offsets from the locations of interest points to the centroid of the object, and the mapping relationship between local appearances of images and their Hough votings were established. The supervised and discriminative codebook was generated, which was used to es- timate the object＇s location via a probabilistic Hough voting, thereby improving the detection precision. Meanwhile, the algorithm reduced the computation and made the detection faster by using SURF local features. Experimental results demonstrated the efficiency of the proposed algorithm.

关 键 词：SURF Hough森林 决策树 训练分类器 概率Hough投票 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]