Analysis of historical photographs
Encadrants : Isabelle Bloch (LIP6), Daniel Foliard (LARCA-UPC), Julien Schuh (Nanterre, Huma-Num), Marie-Eve Bouillon (Archives nationales)
Disponible : NON
Spécialité : IMA/DIGIT
Nombre d'étudiants : 2
Description :
Retrieving the circulation of photographs is very important for historian. This requires finding similar images in different image database, identifying whether an image (or a part of it) has been published in a newspaper, all this taking into account different degradations or printing techniques.
The aim of this project is to contribute to this field.
Pré-requis : Image analysis and machine learning
Travail demandé : 1. Literature review
2. This project can be split into several subprojects, each for one student. In particular one could work on finding an image existing in a database in a large set of newspaper issues (e.g. Le Petit Parisien, available on Gallica). This would allow associating an approximate date and a caption to the image, based on information found on the newspaper, know whether the image has been published several times and when, etc.
Another student could work on simulating potential degradations and printing techniques, such as half-toning, rotogravure, folding, etc. Having a library of such simulation tools can then be used as a data augmentation technique. Preliminary results have shown that this can greatly improve the computation of similarities and the retrieval of similar images.
2. This project can be split into several subprojects, each for one student. In particular one could work on finding an image existing in a database in a large set of newspaper issues (e.g. Le Petit Parisien, available on Gallica). This would allow associating an approximate date and a caption to the image, based on information found on the newspaper, know whether the image has been published several times and when, etc.
Another student could work on simulating potential degradations and printing techniques, such as half-toning, rotogravure, folding, etc. Having a library of such simulation tools can then be used as a data augmentation technique. Preliminary results have shown that this can greatly improve the computation of similarities and the retrieval of similar images.
