Abstract: In some fields of image association rules mining，it is necessary to mine some high confident association rules，which possibly require low support.A method is proposed to mine high confident association rule，while considering support threshold.In order to figure out the problem，the approach uses bSQ（bit Sequential） grid image data format，then rules tree is adopted to avoid the cost of producing large frequent itemsets.Finally，priority in mining association rules from multi-images and image data cube are used to efficiently mine pixel rank association rules from multi-images.The experiment shows the proposed approach could efficiently mine high confident association rules from image data，further validate its feasibility.

Key words: Data Mining, association rules, high confidence, rules tree

摘要： 在图像关联规则挖掘的某些领域，要求提取出具有较高置信度的关联规则，同时对支持度的要求相对较低。提出了一种在兼顾支持度的情况下挖掘出高置信度的图像关联规则的方法。为了便于有效地提取图像关联规则，使用了名为bSQ（bit Sequential）的一种栅格数据格式。而后采取“逐层搜索”的方法，建立规则树，避免了传统方法在处理低支持度时产生的大量频繁项集。最后通过多图像关联规则提取优先级和图像数据立方体等技术在多幅图像中提取基于象素级的关联规则。通过实验证明，该方法能有效地提取图像数据高置信度关联规则，方法具有可行性。

关键词: 数据挖掘, 关联规则, 高置信度, 规则树