A two-step method of searching for similar images in a large image database

Abstract

This article analyzes the methods of searching for similar images in a large database, examines in detail the search for images using perceptual hashes with Hamming distance calculation, using a convolutional neural network and using these methods together. The analysis of the speed and accuracy of the method of searching for similar images by comparing perceptual hashes, the analysis of the accuracy of image classification using a convolutional neural network, the analysis of speed and accuracy when they are used together, including with small changes in images (cropping, rotation, contrast change, sharpness change). According to the results of the study, it was revealed that the joint use of the perceptual hash comparison method and the classification method using a neural network is the most effective.