SpaRClus : Spatial Relationship Pattern-Based Hierarchical Clustering∗

For the past decade, the need of multimedia mining has increased tremendously, especially in image data due to inexpensive digital technologies and fast mounting of image data. In this paper, we, first, show an algorithm, SpIBag (Spatial Item Bag Mining), which discovers frequent spatial patterns in images. Due to the properties of image data, SpIBag considers a bag of items together with a spatial information as a pattern which persists over geometrical transformations, such as scaling, translation, and rotation. Then, based on SpIBag, we propose SpaRClus (Spatial Relationship Pattern-Based Hierarchical Clustering) to cluster image data. Our performance study shows that the method is effective and efficient.