Abstract: Applying the method of bionic pattern，a “matter cognition” based face recognition model is proposed.For the cohesion of a person’s face in certain high-dimensional space，a new method based on the minimum spanning tree of the nearest cover is proposed.Firstly，some human face images are decomposed using wavelet transform，and the decomposition of the four sub-band components is weighted and then fused;secondly，an eigenface-space is extracted using K-L transform;finally，the face is recognized with the method of cover proposed in this paper.The experimental results based on ORL face database show that，compared with traditional pattern recognition，the new algorithm not only solves the misrecognition of none training samples，but also greatly enhances the highly accurate recognition rate.

Key words: biomimetic pattern recognition, minimum spanning tree, Two Dimensional Discrete Wavelet Transform（2D-DWT）, face recognition

摘要： 应用仿生模式识别的理论，提出了基于“认识”事物的人脸识别。根据同一人脸的分布在一定的高维空间的类聚性，提出了一种新颖的基于最小生成树的最邻近覆盖方法。利用小波变换对人脸图像进行分解，对分解后的4个子带分量进行加权融合，用K-L变换提取人脸特征子空间，应用提出的覆盖方法进行人脸识别。在ORL 人脸库上的实验结果表明，与传统的模式识别相比，不仅解决了未训练样本的误识问题，而且能获得更高的识别率。

关键词: 仿生模式识别, 最小生成树, 二维离散小波变换, 人脸识别