查询结果:   梁洋洋,陈宇,杨健.基于深度自编码器网络的人脸特征点定位方法[J].计算机应用与软件,2016,33(9):139 - 142.
中文标题
基于深度自编码器网络的人脸特征点定位方法
发表栏目
人工智能与识别
摘要点击数
724
英文标题
FACIAL LANDMARK LOCALISATION APPROACH BASED ON DEEP AUTOENCODER NETWORKS
作 者
梁洋洋 陈宇 杨健 Liang Yangyang Chen Yu Yang Jian
作者单位
南京理工大学计算机科学与工程学院 江苏 南京 210094     
英文单位
School of Computer Science and Engineering, Nanjing University of Science and Technology, Nanjing 210094, Jiangsu, China     
关键词
人脸特征点定位 深度学习 自编码器网络 逐步求精
Keywords
Facial landmark localisation Deep learning Autoencoder networks Coarse-to-fine
基金项目
国家自然科学基金面上项目(61472187)
作者资料
梁洋洋, 硕士生,主研领域:人脸识别。陈宇,博士生。杨健,教授。 。
文章摘要
使用深度学习网络技术的人脸特征点定位方法已经取得了比较突出的效果。然而,人脸图像由于姿态、表情、光照、遮挡等变化而具有复杂多样性,因此数目较多的人脸特征点(超过50个特征点)定位依然有很大的挑战性。设计了三层级联的自编码器网络,并通过由粗到精的方法对多数目的人脸特征点进行定位。第一层网络以整张人脸图像为输入,直接估计人脸轮廓和部件位置,从而将特征点分成三部分(眼眉鼻,嘴巴和人脸轮廓)进行下一步定位;之后的两层网络分别对各部件特征点进行估计求精。在LFPW、HELEN数据库上的实验表明,该方法能够提高人脸特征点定位的准确性和鲁棒性。
Abstract
Facial landmarks localisation methods using deep learning network technology have achieved prominent effect. However, the localisation of larger number of facial landmarks (more than 50 points) still have lots of challenges due to the complex diversities in face images caused by pose, expression, illumination and occlusion, etc. This paper designs a three-level cascaded autoencoder network, which are employed to locate a large number of facial landmarks in a coarse-to-fine manner. The first level of the network estimates facial contour and component positions directly by tacking the whole face image as input, which divides landmarks into three parts (eyes and nose, mouth, and facial contour) for the next localisation steps; the following two level of the network estimate and refine the landmarks of each part respectively. Experiments conducted on LFPW, HELEN databases show that the approach can improve the accuracy and robustness of facial landmark localisation.
下载PDF全文