摘要
将关联规则挖掘算法推广到图像标注领域,提出了适用于图像语义标注任务的加权关联规则挖掘算法。通过为每个标签及标签集合赋予一定权重,可以保留出现次数少却具有重要意义的标签,以更好地挖掘语义标签之间潜在的有价值的规则。对语义概念之间的层次关系进行了研究,利用高层语义概念对图像标签的结果集合进行扩展,以避免人工标注过程中的不完整标注和遗漏标注问题。实验验证表明,该算法在发现关联规则的数量和扩展标签的质量上性能都优于经典的Apriori算法,证明了该算法的有效性。
This paper generalizes association rule mining algorithm to the field of image auto--annotation and put forward a weighted association rule mining algorithm which is applied to image auto--annotation. By allocating each label and label set a certain weight, label which appears less but has great significance was reserved in order to mine potential and valuable rules between semantic label better. This paper also studies the hierarchical relation between semantic concept, using high--level semantic concept to expand image labels set, which avoid the problem of incompleteness and omission in the process of manual annotation.
出处
《软件导刊》
2016年第10期130-133,共4页
Software Guide
基金
北京市自然科学基金重点项目(KZ201410011014)
北京市教委科研计划面上项目(KM201510011009
KM201510011010)
关键词
语义标签关联关系
图像标注
加权支持度
加权置信度
语义概念分层
Semantic Tags Association
Image Annotation
Weighted Support
Weighted Confidence
Semantic Concept Hierarchy
