摘要
针对现实生活中采集的特定个体数据稀疏而导致学习时产生过拟合的问题,提出了一种基于迁移学习的特定个体学习模型。首先,利用直推式迁移学习从源数据提取有用信息并将其添加到目标数据;然后,利用核主成分分析进行特征提取;最后,将各个样本建模为流形,并利用稀疏系数重建和k近邻分类器完成识别。在PAINFUL数据库上的实验结果表明,所提模型在痛苦表情上的识别精度可高达96.29%,ROC下方面积可高达0.904,相比其他几种较新的模型,所提模型取得了更好的识别性能。
For the issue that specific person data searched from real - world is sparse which will cause fitting problem in learning, a specific person learning model based on transductive transfer learning （TTL） is proposed. Firstly, TTL learning is used to extract useful information from original data and add them into objective data. Then, KPCA is used to extract feature. Finally, all the samples are modeled be manifolds, sparse coefficient reconstruction and k neighbor classifier is used to finish face recognition. Experimental results show that the recognition accuracy of proposed model can arrive at 96.29% and the AUC of ROC can arrive at 0.904. Proposed model has better recognition performance than several other advanced models.
出处
《电视技术》
北大核心
2015年第21期99-103,共5页
Video Engineering
基金
国家自然科学基金项目(61103143)
河南省科技厅科技发展计划项目(134300510037)
平顶山学院青年科研基金项目(PDSU-QNJJ-2013010)
关键词
直推式迁移学习
面部表情识别
核主成分分析
特定个体学习
稀疏系数重建
transductive transfer learning
facial expression recognition
kernel principal component analysis
specific person learning
sparse coefficient reconstruction
