Image collections clustering in large databases on the basis of recurrent optimization

Abstract

The following paper considers methods for clustering large amounts of data and proposes a modification of the density-based approach to clustering multimedia objects with disturbance. The analysis of the existing DENCLUE method is carried out, and the matrix influence function is introduced, which makes it possible to effectively use this approach in the analysis of multidimensional objects, the collections of images, video and multimedia data in particular. The introduced matrix form makes it possible to increase the speed of clustering due to the absence of vectorization-devectorization of the initial data.