Journal of Computer Applications ›› 2017, Vol. 37 ›› Issue (11): 3226-3230.DOI: 10.11772/j.issn.1001-9081.2017.11.3226

Previous Articles     Next Articles

Super-resolution and frontalization in unconstrained face images

SUN Qiang, TAN Xiaoyang   

  1. College of Computer Science and Technology, Nanjing University of Aeronautics and Astronautic, Nanjing Jiangsu 210016, China
  • Received:2017-05-11 Revised:2017-06-27 Online:2017-11-10 Published:2017-11-11
  • Supported by:
    This work is partially supported by the National Natural Science Foundation of China (61373060,61672280)

低质量无约束人脸图像下的超分辨率摆正

孙强, 谭晓阳   

  1. 南京航空航天大学 计算机科学与技术学院, 南京 210016
  • 通讯作者: 孙强
  • 作者简介:孙强(1992-),男,陕西西安人,硕士研究生,主要研究方向:计算机视觉、人脸识别;谭晓阳(1971-),男,重庆人,教授,博士,CCF会员,主要研究方向:计算机视觉、模式识别、机器学习。
  • 基金资助:
    国家自然科学基金资助项目(61373060,61672280);青蓝工程。

Abstract: Concerning the problem that face recognition is affected by the factors such as attitude, occlusion, resolution and so on, a method for image super-resolution and face frontalization in unconstrained image was proposed, which could generate high-quality and standard front images. The projection matrix between the input image and 3D model was estimated to generate the standard front image. Also, through the characteristics of face symmetry, the missing pixels by occlusion and attitude could be filled. In order to avoid the loss of pixel information during the process of generating standard front image and improve the image quality, a deeply-recursive convolutional network which had 16 layers was introduced for image super-resolution. To ease the difficulty of training, two extensions were proposed:recursive-supervision and skip-connection. The experimental results on the processed LFW datasets show that it is surprisingly effective when used for face recognition and gender estimation.

Key words: face recognition, face frontalization, 3D reconstruction, image super-resolution, deeply-recursive convolutional network

摘要: 针对人脸识别算法准确率受面部姿态、遮挡、图像分辨率等因素影响的问题,提出一种超分辨率摆正的方法,作用于低质量无约束输入图像上,生成高清晰度标准正面视图。主要通过估计输入图像与3D模型间的投影矩阵,产生标准正面视图,通过人脸对称性的特点,补全由于姿态、遮挡等原因所产生的面部缺失像素。在摆正过程中,为了提高图像分辨率以及避免面部像素信息丢失,引入一个16层的深度递归卷积神经网络进行超分辨率重构;并提出两个扩展:递归监督和跳跃链接,来降低网络训练难度以及缩小模型体量。在经过处理的LFW数据集上实验表明,该方法对人脸识别和性别检测算法的性能具有显著提升作用。

关键词: 人脸识别, 人脸摆正, 3D重建, 超分辨率重构, 深度递归卷积神经网络

CLC Number: