The article considers the currently existing methods of image search and retrieval in video databases. It is proposed to build the database architecture using the algebra of complete and closed groups as the elements of the active perception theory developed at the department of Computer Systems and Technologies of the R.E. Alekseev Nizhni Novgorod State Technical University. The developed algorithms can find not only original, but also noisy, edited, and corrupted images. We have also developed a method to search for similar images that can compete with existing image retrieval techniques on the Internet in terms of reliability and performance. The experimental results obtained for a database of arbitrary images with about 2 million objects are presented.
|
