Content-based Image Search Web Application

This paper is dedicated to content-based image retrieval (CBIR). Possible applications of CBIR techniques, generic steps of a CBIR algorithm are presented, the mathematical models implemented in this work are described along with possible metrics for determining image similarity based on those models. Besides, some approaches to comparing retrieval accuracy of CBIR systems and databases of annotated imagery used for benchmarking are discussed. The series of benchmarking experiments on the methods implemented in this work are described and their results outlined. The aim of the work is to develop a Web-application for content-based image retrieval. Based on the results of benchmarking experiments, the two most accurate approaches are: a method based on Google Vision API service, color histograms and k-means clustering, and a method based on Gabor filters and color histograms. By calculating the weight of features provided by Google Vision API dynamically, a significant improvement in retrieval accuracy was achieved by incorporating other mathematical models into the CBIR algorithm. The paper contains 47 pages, 3 chapters, 15 illustrations, 1 table, 41 bibliography items, 6 appendices. Keywords: content-based image retrieval, CBIR, Web-application, comparative analysis, Google Vision API, Gabor Filter, Color Histogram, Perceptual Hash.