Development of a Search System for Similar Images

Abstract

This paper discusses the search for images based on their content. Measuring visual similarity between two or more instances is a fundamental task of computer vision. For a person, evaluating the similarity of images is a natural task. However, in computer vision this is a complex problem that arises mainly due to the semantic gap. This paper discusses the use of a neural network, which allows you to get an estimate of the similarity between a pair of visual representations extracted from input data. The difference between the hashes of the two images obtained at the output of the neural network allows us to evaluate their similarity. The results of the developed algorithm and its software implementation are compared with existing solutions and show the best result on various data sets.