Similarity search in fuzzy object databases

Fuzzy object databases are becoming more and more important in the context of image analysis. Examples include satellite images where blurred trees, houses or lakes can still be organized and searched in a meaningful manner and biomedical images which can be utilized to find similar disease patterns and monitor disease progress. One problem of the underlying data is that it contains blurred image content, i.e., fuzzy data. Therefore, an image-based similarity search, which can process huge amounts of fuzzy data in an efficient and effective way, is desirable. The aim of this work is to develop efficient and effective methods for similarity search in fuzzy object databases. First, a suitable similarity measure based on a shape similarity is proposed. Based on this, two novel k-nearest neighbor algorithms for efficient similarity search are presented. The first approach gains efficiency at the cost of incurring only approximate results, while the second approach uses a filter-refinement approach to prune computation. Our experimental evaluation shows the efficiency of the proposed algorithms.