Abstract:
Small sample problem may influence the performance of Bayesian learning algorithms. To overcome the drawback, this paper presents a Bayesian relevance feedback algorithm based on semi-supervised learning. It labels the unlabeled samples by learning the labeled information by users, which enhance the learning ability of the relevance feedback algorithm. Experimental results show that the algorithm achieves better performance than traditional Bayesian learning algorithm.
Key words:
video retrieval,
relevance feedback,
semi-supervised learning,
Bayesian learner,
unlabeled sample
摘要: 小样本问题会制约贝叶斯相关反馈算法的学习能力。为此,提出一种基于半监督学习的视频检索贝叶斯相关反馈算法,其中一个分类器用于估计视频库中每一个镜头属于目标镜头的概率,另一个半监督学习分类器用于判断用户未标记镜头是否与目标相关,由此扩大贝叶斯学习器的训练数据集,提高其分类能力。实验结果表明,该算法提高了贝叶斯算法的检索性能。
关键词:
视频检索,
相关反馈,
半监督学习,
贝叶斯学习器,
未标记样本
CLC Number:
DENG Li, JIN Li-Zuo, BI Min-Dui. Relevance Feedback Algorithm for Video Retrieval Based on Semi-supervised Learning[J]. Computer Engineering, 2011, 37(22): 281-283.
邓丽, 金立左, 费敏锐. 基于半监督学习的视频检索相关反馈算法[J]. 计算机工程, 2011, 37(22): 281-283.