An Efficient Signature Extraction Method for Image Similarity Retrieval

The goal of similarity retrieval is to retrieve images that are similar to query image. Good access methods for large image databases are very important for efficient retrieval. The 2D B-string-based and unique-ID-based signature methods can provide four kinds of similarity retrieval, object and type-i, where 0 ≤ i ≤ 2, and can distinguish 169 spatial relationships. However, 169 spatial relationships are still not sufficient to show all kinds of spatial relationships between any two objects in 2D space, such as directional relationships, like north. Moreover, most of the previous similarity retrieval methods, for the sake of simplicity, apply the MBRs of two objects to define the spatial relationship between them. The topological relationships, however, between objects can be quite different from the spatial relationship between their respective MBRs. Therefore, in this paper, we propose a new method that focuses on the above two problems. To solve the first problem, we add 9 directional relationships to the 169 spatial relationships. In this way, we can distinguish up to 289 spatial relationships in 2D space. To handle the second problem, we apply the concept of topological relationships in our proposed method. Based on the results of our simulation study, we show that our method can achieve a higher correct match rate than the 2D B-string-based and unique-ID-based signature methods can.