摘要
人脸图像的年龄估计关键步骤在于提取特征,但各种复杂因素的影响诸如生长地域、生活习惯、遗传基因等,均会使得最终的估计结果值产生偏差,为此提出一种双特征融合模型来刻画人脸由于年龄变化而产生的不同特征,将梯度直方图(HOG)和局部二值模式(LBP)共同作为特征描述算子进行特征提取,再将二者进行特征融合,构成更为精确的年龄估计模型,随后采用支持向量回归的方法得到年龄回归函数.将所提实验在人脸数据集上,结果表明该模型可以快速且较为准确地对人脸图像进行年龄估计,较之于单独提取HOG特征,可将年龄估计的平均误差缩小0.7岁,较之于单独提取LBP特征可将年龄估计的平均误差缩小1.2岁,且该融合模型对幼儿期至青少年期的年龄估计表现效果最佳,平均估计误差在3.91岁;成年至中年期效果其次,平均估计误差在4.41岁;老年阶段人脸的平均估计误差为5.80岁.
The most critical step in estimating the age based on face images is to extract features.The appearance of the face is affected by various complicated factors such as growing regions,living habits,genetics,etc.,which will cause deviations in the final estimated result values.In this paper,a dual feature fusion model is proposed to characterize the different features of human faces due to age changes.The histogram of oriented gradient(HOG)and the local binary pattern(LBP)are used together as feature description operators for feature extraction.Then the two features are combined to form a more accurate age estimation model.The age regression function is obtained using the support vector regression(SVR)method.Finally,the model proposed is put into experiment on the face dataset.The results show that the model can estimate the age of the face image quickly and accurately.Compared with the HOG feature alone,the average error in the age estimation can be reduced to 0.7 years.The average error of age estimates can be reduced by 1.2 years compared with that of extracting LBP features alone.Besides,compared with two representative deep learning methods,it performs equally well.The fusion model has the best performance in estimating the age from early childhood to adolescence,with an average estimated error of 3.91 years;the effect for the adult to middle age is the second,the average estimated error being 4.41 years;the average estimated error of the elderly face is 5.80 years old.
作者
刘庆华
李智
LIU Qinghua;LI Zhi(School of Computer Science and Engineering, Jiangsu University of Science and Technology, Zhenjiang 212100, China)
出处
《江苏科技大学学报（自然科学版）》
CAS
北大核心
2021年第3期50-55,共6页
Journal of Jiangsu University of Science and Technology:Natural Science Edition
关键词
年龄估计
梯度直方图
局部二值模式
支持向量回归
age estimation
histogram of oriented gradient
local binary pattern
support vector regression
