Development of a multidimensional data model for efficient content-based image retrieval in big data storage

The object of the study is content-based image retrieval. The subject of the study is the models and methods of content-based image retrieval in Big Data storage under high-intensity search queries. The purpose of this study is to develop a multidimensional data model and related search methods that can use and adapt to existing image descriptors and perform searches based on them. The task is to: analyze modern approaches and solutions for effective content-based image retrieval, formulate the problem and requirements for the search system; develop a model that will effectively process descriptors and place them inside in such a way as to minimize the number of descriptors with which comparisons need to be made during the search; develop a search algorithm; develop metrics, perform experiments and compare the results obtained with analogs. The methodology includes analyzing the search process and highlighting the stages of descriptor formation, its placement in the model, determining the level of similarity and comparing and forming the results; building a data model and placing it in memory; conducting experiments with data sets available on the Internet; evaluating the effectiveness of the search and forming the resulting tables for comparison with analogs. The following results were obtained: Multi-Dimensional Cube (MDC) model with optimizations and search algorithms was developed. It was compared with the brute-force search and the search that uses Inverted Multi-Index (IMI). The experimental results showed that MDC provides the best search speed among competitors. Demonstrates search quality at the level of competitors. The search labor intensity shown by the MDC is the best for searching for original images in the storage (checking whether they are present in storage). The labor intensity of searching for modifications of the images is better than in brute-force search by more than 100 times, but worse by 30% than when using IMI. Conclusions: The developed MDC model with its search algorithm solves the task of efficient content-based image retrieval, using existing image descriptors. The obtained results are satisfactory, but a promising direction is to improve the cell boundaries optimization algorithm and apply parallel computing.