In this paper, we propose a new binary pixel matching method to
measure the similarity of two objects based on the overlapped
areas in a data warehouse. In the proposed approach, suspicious
objects are extracted as binary images that consist of background
and foreground, which are compared based on counting
overlapped pixels. Before measuring the exact similarity between
two objects, we calculate a difference in the shape index, perform
rotation normalization, then, estimate accurate similarity, if the
difference in the shape index is less than 50%. The results
demonstrate that the proposed approach can significantly improve
both the accuracy of similarity measurement and the performance
of computational speed, when compared with traditional content
based image retrieval method approaches that measure the
similarity based on contours.