Using Spatial Reasoning in CPM Based Image Retrieval

Abstract

Traditional image retrieval methods are based on the spatial relationship regarding objects as centroids, and take the advantage of point based spatial relational models to measure the similarity of images. They can not distinguish the directional or topological relationships, so it will return irrelative results. We give a new image retrieval algorithm RACPM by using spatial reasoning and common pattern method. The RACPM compares the rectangle algebra expression of the objects both in the query image and the database image on the x axis and the y axis, and these two pairs are “similar” if and only if all the expressions of them are identical respectively. Compared with the original CPM, the experimental results show that the RACPM outperforms CPM, not only on the time consuming, but also provides more precise retrieval results.