CBL - Campus del Baix Llobregat

Projecte llegit

Títol: Relate that image!: A tool for finding related cultural heritage images.

Estudiants que han llegit aquest projecte:

Director/a: MARINESCU, MARIA-CRISTINA

Departament: DAC

Títol: Relate that image!: A tool for finding related cultural heritage images.

Data inici oferta: 22-02-2019     Data finalització oferta: 22-10-2019


Estudis d'assignació del projecte:
    MU MASTEAM 2015
Tipus: Individual
 
Lloc de realització: EETAC
 
Paraules clau:
data mining, natural language processing, cultural heritage, image
 
Descripció del contingut i pla d'activitats:
The main goal of this project is to identify related cultural heritage images based on any of several numbers of similarity criteria, by applying data mining techniques.

Pictures can be related not only via direct image analysis - which is not the goal of this project- but by semantic similarity of image descriptions, tags and other metadata, style, author relationships, etc.

The main objectives are:

- Study and analyze the state of the art, and the different data mining and natural language processing tools that can be useful to find relations between paintings.

- Discover different possible similarity criteria and implements methods to retrieve images related to an initial image from the domain of cultural heritage - in particular paintings.

- Develop a user interface that takes an image and returns related paintings.

- Optionally, create training corpuses.
 
Overview (resum en anglès):

Museums, galleries, art centers, etc. are increasingly seeing the benefits of digitalizing their art work collections – and acting on it. The more visible benefits usually have to do with advertising, involving the citizens, or creating interactive tools that get people interested in coming to museums or buying art. With the availability of these increasingly large collections, analysis of art images has gained attention from researchers.

This master thesis proposes a tool to recommend paintings that are similar to a given image of an artwork. We define different similarity measures that include criteria existent in the metadata associated with the digitized pictures (e.g. style, genre, artist, etc.), but also image content similarity. The work is more closely related to existing approaches on automatic classification of paintings, but also shares techniques with other areas such as image clustering.

Our goal is to offer a tool that can enable creative uses, support the work of gallery / museum curators, help create interesting and interactive educational content, or create clusters of images as training sets for further learning and analysis algorithms.

© CBLTIC Campus del Baix Llobregat - UPC