Novel visual information indexing in relational databases

This paper presents a novel relational database architecture aimed for visual objects classification and retrieval based on the content of images. Most methods search and classify images based on their content use algorithms for generating and comparing keypoint descriptors. We present a new method to quickly compare these visual features, based on fuzzy sets, which is embedded in a real database environment. We also present a full framework as an extension of Microsoft SQL Server database, which allows to search and classify images directly from SQL query command in large data sets of visual objects. The system utilizes standard indexing mechanisms and SQL queries. We tested the method on a state-of-the-art dataset.