摘要
针对级联回归模型依赖形状初始化且结构复杂使其在人脸特征点定位中速度慢、精度低的问题,提出了改进的级联回归人脸特征点定位算法。采用仿射变换参数回归初始化人脸形状,使变换后的初始形状更接近真实人脸以提高模型的收敛速度和精度;在各特征点局部区域构造随机蕨局部学习器,并学习得到易于计算且高度稀疏的二值化特征应用提高模型的速度;对二值化特征使用全局线性回归求得形状增量,实现特征点定位。仿真实验结果表明:相比于原算法,所提算法在LFPW,HELEN,AFW库上定位误差平均降低了11%,定位时间平均减少了14%。
Aiming at problem of low precision and slow running of cascade regression model in locating face alignment because of dependent shape initialization and high complexity,an improved cascade regression face alignment algorithm is proposed.Face shape is initialized by affine transformation parameter regression,and initial shape of the transformed shape is closer to the ground truth for improving model accuracy and convergence speed.Random ferns that follows local learning principle are constructed in local area of each feature point,and learn to obtain easily and highly sparse binary feature to improve speed of model. Global linear regression is used to compute binary feature and obtain shape increment to complete face alignment. Experimental simulation results show that average error of face alignment reduces 11 %,and average cost of time decreases 14 % compared with original algorithm on LFPW,HELEN and AFW datasets.
作者
贾项南
于凤芹
杨慧中
陈莹
JIA Xiang-nan;YU Feng-qin;YANG Hui-zhong;CHEN Ying(School of Internet of Things Engineering, Jiangnan University, Wuxi 214122, China)
出处
《传感器与微系统》
CSCD
2018年第4期58-61,共4页
Transducer and Microsystem Technologies
基金
国家自然科学基金资助项目(61573168)
中央高校基本科研业务费专项资金资助项目(JUSRP51733B)
关键词
级联回归模型
人脸特征点定位
仿射变换
初始化
随机蕨
全局线性回归
cascaded regression model
face feature point alignment
affine transformation
initialization
random fern
global linear regression
