Aplicatiu per la base de dades de Marques d’impressors per tal de poder detectar imatges similars de marques

Abstract

[en] This project is based on the creation of an application that allows the end user to consult the database of images of printer’s brands in charge of the Reserve Library of the University of Barcelona and to see images similar to a brand. For this, different methods are used to extract characteristics from the image, then techniques to improve the results obtained and finally, a web development that allows to carry out all the search functionalities. The project is divided into two phases, one of comparison and research and another part of software development for the creation of the application. Throughout the project we will see different technologies used for both the comparison part and for the development part of the final web. In the research and comparison we will see how to extract the descriptive vectors of different models and then compare them to obtain K similar images, which is currently called Transfer learning. Then we will apply a fine tuning to the selected model to improve the results. Once we have this model located we will go to the development of the final application in which consists of two parts: an API using the Django framework to facilitate the work and then the Frontend using the framework of Angular which will serve to show all the images and printers, always taking into account a satisfactory user experience. During the whole project we will experiment with different things that we will see in order to obtain the results we want at the end of the project.