计算机工程与应用 ›› 2018, Vol. 54 ›› Issue (24): 116-122.DOI: 10.3778/j.issn.1002-8331.1805-0117

• 模式识别与人工智能 • 上一篇    下一篇

类矩阵神经核特征融合的人脸识别方法

臧海娟   

  1. 江苏理工学院 计算机工程学院,江苏 常州 213001
  • 出版日期:2018-12-15 发布日期:2018-12-14

Face recognition method based on feature fusion of quasi-matrix neural kernel

ZANG Haijuan   

  1. School of Computer Engineering, Jiangsu University of Technology, Changzhou, Jiangsu 213001, China
  • Online:2018-12-15 Published:2018-12-14

摘要: 针对人脸识别过程中所提取特征向量的信息不完整性与整体图像信息数据量较大的问题,提出一种类矩阵神经核特征融合的人脸识别方法。该方法为深度神经网络的首层升维操作,首先将人脸数据作为特征向量的集合,利用随机矩阵列采样构成随机特征矩阵;其次设计深度神经核将随机特征矩阵映射为高维空间中的新特征向量;最后利用快速收缩算法求解匹配过程中的不定线性代数方程组,使收敛速度达到二阶收敛。该方法既克服了直接使用人脸图像数据空间复杂度较大的问题,又增加了特征的非线性结构,提高了特征向量的表达能力。实验结果表明,该方法识别率高、稳定性强、鲁棒性好,适合处理大型数据。

关键词: 特征融合, 深度核, 快速收缩迭代, 稀疏表达

Abstract: Considering the information of face feature vector is incomplete and the whole image content has great data amount, this paper presents a method of face recognition based on feature fusion of quasi-matrix neural kernel. This method can be interpreted as the first level lifting operation of deep neural network. Firstly, face data is used as a set of feature vectors, and random matrix sampling is used to form a random feature matrix. Secondly, the stochastic matrix can be mapped to a new feature vectors in a high dimensional space based on deep neural core designed in this paper. The new feature not only overcomes the problem of huge space complexity, but also increases the nonlinear structural characteristics and improves the ability of expression. Finally, a fast iterative shrinkage-thresholding algorithm is proposed to find solution, with the convergence speed is quadratic convergence. Experimental results show that the proposed method has high recognition rate, strong stability and good robustness, is well-suited for processing large data.

Key words: feature fusion, deep neural core, fast iterative shrinkage-thresholding, sparse representation