Indirect Matching Using Dynamic Cues for Effective Similar Shape Retrieval

Shape images such as vector graphics have become popular for creating various types of online content. The importance of similarity search systems for shape images has been increasing. As a fast similarity retrieval method, indirect matching has been proposed. The indirect matching quickly estimates the similarity between each shape in the database and the query using a small number of cue images. The retrieval accuracy of the indirect matching, however, significantly changes depending on the combination of cue images. In addition, we have found the best combination of cue images is dependent on the query. To improve the retrieval accuracy of the indirect matching, we propose a dynamic selection method to determine optimal cue images depending on a given query in real time. Experimental results have shown that retrieval accuracy and performance can be improved by the proposed method.