基于无监督方法的视频中的人物识别

基于无监督的显著性学习方法提出一种新颖的人物识别方法。它在训练程序部分不需要身份标签就能提取出突出的特征。首先利用相邻约束斑块匹配在图片对之间构建稠密对应。该方法在处理由于较大的视觉角度变化和人物姿势变化而引起的图片对之间不对应的情况非常有效。其次，它应用一种无监督的方法来学习人物的显著性。为了提高实验的性能，在斑块匹配过程中融合了这种人物的显著性特征。在VIPeR数据集上进行的实验证实了该方法的正确性，且性能略优于文献中提出的eBiCov方法及eLDFV方法。

Person Recognition in Video Based on Unsupervised Method

In this paper , we propose a novel method for person recognition based on a salient learning with unsupervised .The sa-lient features can be extracted without the person labels in the training procedure .First, we utilize patch matching with adjacency constrained to create dense correspondence between image pairs and it shows validity in processing misalignment problem caused by larger viewpoint and pose changes .Second, we learn person salience by applying an unsupervised approach .In order to im-prove the performance of experiment , the person salience is combined with patch matching .This approach on the VIPeR dataset is proved the effect is validated .And the performance is better than the eBiCov method and the eLDFV method .