¡¡Chinese Journal of Computers   Full Text
  TitleConstructive Neuron Networks Classification Algorithm Based on Biomimetic Pattern Recognition
  AuthorsWANG Xian-Bao1),2) ZHOU De-Long1) WANG Shou-Jue2),3)
  Address1)(College of Information Engineering, Zhejiang University of Technology, Hangzhou 310014)
2)(Institute of Intelligent Information System, Zhejiang University of Technology, Hangzhou 310014)
3)(Laboratory of Artificial Neural Networks, Institute of Semiconductors, Chinese Academy of Sciences, Beijing 100083)
  Year2007
  IssueNo.12(2109¡ª2114)
  Abstract &
  Background 		Abstract In this paper, based on Biomimetic Pattern Recognition theory, a Constructive Neuron Networks Classification Algorithm is proposed. The theory believes that: There exists at least one gradual change course between two things and all the things in this gradual change course belong to the same class, if these two things belong to the same class. Analyzing the geometry meaning of different structure neurons in high dimensional space, high dimensional geometrical distribution of the sample set in the feature space can be covered by constructing a new type of ANN. The high recognition rate of the double screw curves experiment has proved the validity of the new algorithm.

keywords pattern recognition; neuron; neural networks; constructive neuron networks; geometrical distribution; high dimensional space

background As an important research field of artificial intelligence, pattern recognition has made a rapid development in recent decades. Its various theories and methods were applied widely in other science and technology fields. Since being proposed, biomimetic pattern recognition has been applied in practicality object recognition, human face personal identification and human face recognition and the results have showed it precede traditional pattern recognition. Based on biomimetic pattern recognition, analyzing the geometry meaning of different structures of neuron in high dimensional space, a new type of NN was built, which used two structure neurons. It broke through the limitation that built neural networks only using single structure of neuron.
The topic of this paper is a part of "High Dimensional Descriptive Geometry Biomimetic Informatics". The theory was developed by professor Wang Shou-Jue.
This work is supported by the National Natural Science Foundation of China under grand No.60572077, which title is "Human Face Recognition Research Based on Image Fusion".