• 中国计算机学会会刊
  • 中国科技核心期刊
  • 中文核心期刊

计算机工程与科学 ›› 2022, Vol. 44 ›› Issue (02): 291-297.

• 图形与图像 • 上一篇    下一篇

改进标签一致KSVD字典学习的人脸识别算法

严春满,张昱瑶,张迪   

  1. (西北师范大学物理与电子工程学院,甘肃 兰州 730070) 
  • 收稿日期:2020-09-25 修回日期:2020-12-07 接受日期:2022-02-25 出版日期:2022-02-25 发布日期:2022-02-18
  • 基金资助:
    国家自然科学基金(61861041)

Face recognition based on improved label consistent KSVD dictionary learning

YAN Chun-man,ZHANG Yu-yao,ZHANG Di   

  1. (College of Physics and Electronic Engineering,Northwest Normal University,Lanzhou 730070,China)
  • Received:2020-09-25 Revised:2020-12-07 Accepted:2022-02-25 Online:2022-02-25 Published:2022-02-18

摘要: 针对噪声污染、光照变化等复杂环境下人脸图像识别问题,提出一种改进标签一致KSVD字典学习的人脸识别算法。该算法通过改变标签一致KSVD算法的字典更新方式,用主成分分析算法分解误差项,用最大特征值对应的特征向量修改字典原子。通过字典学习过程得到原子与类别标签对应的判别性字典。目标函数综合了重建误差、稀疏编码误差和分类误差。最后,在分类阶段利用学习到的字典和分类器参数对测试样本进行分类。在有光照变化的Extend Yale B人脸库、表情变化以及遮挡影响的AR人脸库上分别取得了99.01%和97.94%的平均识别率。同时,在有噪声存在的情况下,该算法具有较好的鲁棒性。

关键词: 人脸识别;标签一致KSVD算法, 字典学习, 主成分分析

Abstract: Aiming at the problem of face recognition under complicated conditions such as noise pollution and illumination changes, 
a face recognition algorithm based on improved label consistent KSVD dictionary learning is proposed. The algorithm changes the dictionary update way of the label consistent KSVD algorithm, uses the principal component analysis algorithm to decompose the error term, and the dictionary atom is modified by the eigenvector corresponding to the maximum eigenvalue. Then, through the dictionary learning process, the discriminative dictionary corresponding to atoms and category labels is obtained. The objective function combines reconstruction error, sparse coding error and classification error. Finally, in the classification stage, the learned dictionary and classifier parameters are used to classify the test samples. The average recognition rates of 99.01% and 97.94% are obtained on extended Yale B database with illumination conditions, and AR database with various facial expressions and occlusion effect, respectively. At the same time, the algorithm is more robust in the presence of noise. 


Key words: face recognition;label consistent KSVD algorithm;dictionary learning, principal component analysis