• 中国计算机学会会刊
  • 中国科技核心期刊
  • 中文核心期刊

J4 ›› 2013, Vol. 35 ›› Issue (4): 104-110.

• 论文 • 上一篇    下一篇

基于稀疏信号同伦形变的鲁棒性特征学习方法

惠寅华,李凡长   

  1. (苏州大学计算机科学与技术学院,江苏 苏州 215006)
  • 收稿日期:2012-05-20 修回日期:2012-10-13 出版日期:2013-04-25 发布日期:2013-04-25

Homotopybased sparse coding for robust feature learning    

 HUI Yinhua,LI Fanzhang   

  1. (School of Computer Science and Technology,Soochow University,Suzhou 215006,China)
  • Received:2012-05-20 Revised:2012-10-13 Online:2013-04-25 Published:2013-04-25

摘要:

在人脸识别问题中,为克服同一个人由于在表情与姿势上的不同给识别带来的困难,设计了基于同伦伪不变性的目标识别与图像检索方法。人脸识别中的光照变化问题也可以看作是由于人脸在空间中相对图像采集设备连续旋转平移所造成的。为了在采集的样本上提取这一同伦等价特征,结合稀疏表示提出了一个更一般的鲁棒性特征学习方法,并在Yale B数据集上进行了测试,得到了不错的效果。

关键词: 同伦等价, 稀疏表示, 形变, 关键点, 双射, 能量函数

Abstract:

In the face recognition problem, in order to overcome the difficulties caused by the changes in posture and facial expressions, homotopic image pseudoinvariants for openset recognition and image retrieval is proposed. However, the illumination change is also caused by the continuous rotation of the face relative to the image acquisition equipment. In this paper we proposed a more general robust feature learning method based on homotopy equivalence with sparse coding. We tested it on the Yale B datasets and got a good result.

Key words: homotopy;sparse coding;morphing;critical_point;discrete bijection;energy function