徐艳.基于样本扩张和最大散度差融合的单样本人脸识别算法[J].光电子激光,2017,28(3):311~315 |
基于样本扩张和最大散度差融合的单样本人脸识别算法 |
Single sample face recognition based on sample augments and maximum scatter diff erence fusion |
投稿时间:2016-01-05 |
DOI: |
中文关键词: 人脸识别 虚拟样本 最大散度差(MSD)鉴别分析 单样本问题 模糊决策 |
英文关键词:face recognition virtual samples maximum scatter difference (MSD) discrim inate analysis single sample problem fuzzy decision |
基金项目:山东省高等学校科研计划(J13LN85)资助项目 (临沂大学 信息学院,山东省网络环境智能计算技术重点实验室 临大研究所,山东 临沂 276005) |
|
摘要点击次数: 850 |
全文下载次数: 0 |
中文摘要: |
为解决只有一个训练样本时最大散度差(MSD) 鉴别分析在人脸识别中的识别性能会降低这一问题,提 出一种基于样本扩张和MSD融合的单样本人脸识别算法。首先,根据人脸的对称相似理论 ,人脸样本的相关变化信息可 以从它的对称脸上提取,并且平均脸也具有要识别测试人脸的某些可能变化,提出组合原始 训练样本及它的虚 拟平均脸和虚拟对称脸作为新的训练样本集;然后,在新的训练样本集上应用类内中间值MS D鉴别分析算法得到最 优投影矩阵,从而可以基于这个最优投影矩阵计算训练样本和待测试样本的特征;最后利用 模糊决策方法进行分类。在ORL和FERET人脸数据库上的大量实验结果表明,本文算法可以提 高识别率,具有一定的鲁棒性。 |
英文摘要: |
When each person has only one training sample,the recognition perform ance of maximum scatter difference (MSD) discriminant analysis in face recognition would be reduced.To solve this problem,a single s ample face recognition algorithm based on sample augments and MSD fusion is proposed in this paper.First,according to the facia l symmetry theory,some relevant information of possible change could be extracted to adapt to the future samples.At the same t ime,the average face may also have some changes of the test faces in the future.So the original training sample,symmetrical virtual f ace and average virtual face are combined to a new training sample set.Then,the MSD algorithm is performed on the new training sample set to get the optimal projection matrix.Therefore,the features of the training sample and testing fa cial images could be obtained by projecting them on to the optimal projection matrix achieved above.During the recognition stage,the fuzzy decision is used to do the classification.Extensive experiment results on famous ORL and FERET show that the algorithm can improve t he recognition rate and has certain robustness. |
查看全文 下载PDF阅读器 |
关闭 |
|
|
|
|
|