Publication
Authentication of Amadeo de Souza-Cardoso Paintings and Drawings With Deep Learning
| Summary: | Art forgery has a long-standing history that can be traced back to the Roman period and has become more rampant as the art market continues prospering. Reports disclosed that uncountable artworks circulating on the art market could be fake. Even some principal art museums and galleries could be exhibiting a good percentage of fake artworks. It is therefore substantially important to conserve cultural heritage, safeguard the interest of both the art market and the artists, as well as the integrity of artists’ legacies. As a result, art authentication has been one of the most researched and well-documented fields due to the ever-growing commercial art market in the past decades. Over the past years, the employment of computer science in the art world has flourished as it continues to stimulate interest in both the art world and the artificial intelligence arena. In particular, the implementation of Artificial Intelligence, namely Deep Learning algorithms and Neural Networks, has proved to be of significance for specialised image analysis. This research encompassed multidisciplinary studies on chemistry, physics, art and computer science. More specifically, the work presents a solution to the problem of authentication of heritage artwork by Amadeo de Souza-Cardoso, namely paintings, through the use of artificial intelligence algorithms. First, an authenticity estimation is obtained based on processing of images through a deep learning model that analyses the brushstroke features of a painting. Iterative, multi-scale analysis of the images is used to cover the entire painting and produce an overall indication of authenticity. Second, a mixed input, deep learning model is proposed to analyse pigments in a painting. This solves the image colour segmentation and pigment classification problem using hyperspectral imagery. The result is used to provide an indication of authenticity based on pigment classification and correlation with chemical data obtained via XRF analysis. Further algorithms developed include a deep learning model that tackles the pigment unmixing problem based on hyperspectral data. Another algorithm is a deep learning model that estimates hyperspectral images from sRGB images. Based on the established algorithms and results obtained, two applications were developed. First, an Augmented Reality mobile application specifically for the visualisation of pigments in the artworks by Amadeo. The mobile application targets the general public, i.e., art enthusiasts, museum visitors, art lovers or art experts. And second, a desktop application with multiple purposes, such as the visualisation of pigments and hyperspectral data. This application is designed for art specialists, i.e., conservators and restorers. Due to the special circumstances of the pandemic, trials on the usage of these applications were only performed within the Department of Conservation and Restoration at NOVA University Lisbon, where both applications received positive feedback. |
|---|---|
| Main Authors: | Chen, Ailin |
| Subject: | Cultural Heritage Authentication Artificial Intelligence Deep Learning Hyperspectral Analysis Visualisation |
| Year: | 2022 |
| Country: | Portugal |
| Document type: | doctoral thesis |
| Access type: | open access |
| Associated institution: | Universidade Nova de Lisboa |
| Language: | English |
| Origin: | Repositório Institucional da UNL |
| Summary: | Art forgery has a long-standing history that can be traced back to the Roman period and has become more rampant as the art market continues prospering. Reports disclosed that uncountable artworks circulating on the art market could be fake. Even some principal art museums and galleries could be exhibiting a good percentage of fake artworks. It is therefore substantially important to conserve cultural heritage, safeguard the interest of both the art market and the artists, as well as the integrity of artists’ legacies. As a result, art authentication has been one of the most researched and well-documented fields due to the ever-growing commercial art market in the past decades. Over the past years, the employment of computer science in the art world has flourished as it continues to stimulate interest in both the art world and the artificial intelligence arena. In particular, the implementation of Artificial Intelligence, namely Deep Learning algorithms and Neural Networks, has proved to be of significance for specialised image analysis. This research encompassed multidisciplinary studies on chemistry, physics, art and computer science. More specifically, the work presents a solution to the problem of authentication of heritage artwork by Amadeo de Souza-Cardoso, namely paintings, through the use of artificial intelligence algorithms. First, an authenticity estimation is obtained based on processing of images through a deep learning model that analyses the brushstroke features of a painting. Iterative, multi-scale analysis of the images is used to cover the entire painting and produce an overall indication of authenticity. Second, a mixed input, deep learning model is proposed to analyse pigments in a painting. This solves the image colour segmentation and pigment classification problem using hyperspectral imagery. The result is used to provide an indication of authenticity based on pigment classification and correlation with chemical data obtained via XRF analysis. Further algorithms developed include a deep learning model that tackles the pigment unmixing problem based on hyperspectral data. Another algorithm is a deep learning model that estimates hyperspectral images from sRGB images. Based on the established algorithms and results obtained, two applications were developed. First, an Augmented Reality mobile application specifically for the visualisation of pigments in the artworks by Amadeo. The mobile application targets the general public, i.e., art enthusiasts, museum visitors, art lovers or art experts. And second, a desktop application with multiple purposes, such as the visualisation of pigments and hyperspectral data. This application is designed for art specialists, i.e., conservators and restorers. Due to the special circumstances of the pandemic, trials on the usage of these applications were only performed within the Department of Conservation and Restoration at NOVA University Lisbon, where both applications received positive feedback. |
|---|