ISSN 1000-1239 CN 11-1777/TP

计算机研究与发展 ›› 2016, Vol. 53 ›› Issue (11): 2623-2629.doi: 10.7544/issn1000-1239.2016.20150630

• 人工智能 • 上一篇    下一篇



  1. (中国科学院半导体研究所 北京 100083) (
  • 出版日期: 2016-11-01
  • 基金资助: 
    国家自然科学基金项目(90920013,61572458);国家公派访问学者项目(201404910237);国家重大科学仪器设备开发专项项目(2014YQ470377) This work was supported by the National Natural Science Foundation of China (90920013,61572458), the China Scholarship Council (201404910237), and the National Key Scientific Instrument and Equipment Development Project (2014YQ470377).

Uncorrelated Locality Preserving Discriminant Analysis Based on Bionics

Ning Xin, Li Weijun , Li Haoguang, Liu Wenjie   

  1. (Institute of Semiconductors, Chinese Academy of Sciences, Beijing 100083)
  • Online: 2016-11-01

摘要: 由于形象思维方式是人类的一种本质思维方式,人类通过各种感官来认知事物的规律性,进而提取出具有代表性的特征,因此通过形象思维的方法来提取事物的本质特征符合人类认知事物的规律.针对人脸识别中特征提取问题,该算法以形象认知规律与无监督判别投影为理论基础,提出了一种仿生不相关空间局部保持鉴别分析(biomimetic uncorrelated locality preserving discriminant analysis, BULPDA)算法.算法首先根据人类形象认知的特性构建了一种新的相似系数表示方法;然后结合不相关空间概念,确保矢量空间具有不相关性;最后给出了基于奇异值分解的矢量空间求解方法,形成了一种特征提取新思路.在标准数据库上的实验结果表明,新算法优于传统的特征提取方法和其他改进的局部保持投影方法.

关键词: 无监督判别投影, 形象认知, 不相关空间, 特征提取, 奇异值分解

Abstract: Imagery thinking model is an essential way of thinking for human being. It cognizes the regularity of things through various human senses, and then extracts the representative features. Therefore, using the method of imagery thinking to extract the essential characteristics of things is in conformity with the law of human cognition. According to the problem of feature extraction in face recognition technology, we propose an uncorrelated space locality preserving discriminant analysis algorithm—BULPDA based on the theory of unsupervised discriminant projection and image cognitive law. On the basis of the characteristics of human image cognitive, the proposed algorithm first builds a new construction method of similarity coefficient. Then, it applies uncorrelated space concepts to ensure the non-relevance of vector space. Finally, it gives the solution of the proposed algorithm based on singular value decomposition. The algorithm presents a new idea of feature extraction. The experimental results on the standard face database show that the proposed algorithm is better than the traditional preserving projection algorithms.

Key words: unsupervised discriminant projection, image cognitive, uncorrelated space, feature extraction, singular value decomposition