摘要
针对稀疏表示用于人脸识别时,由训练样本构成的原字典包含不利于识别的信息,且对测试样本进行编码时,稀疏表示系数判别性不足的问题,提出了一种基于字典学习的局部约束稀疏表示的人脸图像识别方法。该方法首先对原字典,利用Fisher判别的字典学习(FDDL)得到具有字典识别力和编码识别力的结构化字典。然后利用学习得到的字典,对测试样本在局部约束下进行稀疏表示(LCSRC),可以得到更多包含判别信息的有效稀疏表示系数,从而提高整体识别性能。在ORL和AR数据库的实验结果验证了文中方法的有效性。
When sparse representation is used in face recognition, the original dictionary composed of training samples contains in-formation that is not conducive to identification and the sparse representation coefficient for test sample is not discriminative, wepropose a novel face recognition method that integrates the dictionary learning with the locality-constrained sparse representation.This method firstly obtains a structured dictionary with dictionary recognition ability and coding recognition ability by using theFisher Discriminant Dictionary Learning(FDDL) for the original dictionary, Then under the dictionary learning framework, thesparse representation for the test samples with locality constraint(LCSRC) can be used to obtain sparse representation coefficientswith more discriminative information. Thus, the overall recognition performance is improved. Experimental results on the ORL andAR databases verify the effectiveness of the proposed method.
出处
《电脑知识与技术》
2018年第2Z期200-202,共3页
Computer Knowledge and Technology
关键词
稀疏表示分类
Fisher判别字典学习
局部约束
人脸识别
sparse representation classification
fisher discrimination dictionary learning
locality constraint
face recognition