摘要
针对人脸性别识别中单一特征描述子对人脸信息表达的不足,提出一种方向梯度直方图(histogram of oriented gradient,HOG)和多尺度局部二值模式(local binary patterns,LBP)多特征融合的人脸性别识别算法.首先,对输入图像进行裁剪和缩放得到多个分辨率的人脸图像,再分别提取LBP统计直方图并合成一个特征向量;然后,提取目标图像头肩模型的HOG特征得到HOG特征向量;最后,将LBP特征向量与HOG特征向量合成一个新的特征向量,应用支持向量机(support vector machine,SVM)进行训练.在自建的人脸库中采用10折交叉法测试该算法的准确性,准确率可达93.0%,结果表明该算法对人脸性别识别的准确率优于其他单一的特征提取算法.
In order to reduce the limitation of single feature descriptor, the author proposes a feature descriptor joint HOG and multi-scale LBP feature. First, LBP histogram on multi-resolution face region is calculated. Then, the HOG feature is extracted on head-shoulder model. The LBP histo- gram and HOG feature vector are composed into a new feature vector for SVM training finally. Test in self-built libraries uses 10-fold cross-validation. The accuracy rate of 93.0% is obtained. Experi- mental results show this method is superior to the method which just employs single feature de-scriptor.
出处
《扬州大学学报（自然科学版）》
CAS
北大核心
2016年第3期58-62,共5页
Journal of Yangzhou University：Natural Science Edition
基金
广东省自然科学基金资助项目(2016A030310077)
关键词
人脸性别识别
方向梯度直方图
多尺度局部二值模式
支持向量机
face gender recognition
histogram of oriented gradient
multi-scale local binary pat-terns
support veetor machine
