Image processing and machine learning for art investigation
Lead Research Organisation:
Imperial College London
Department Name: Electrical and Electronic Engineering
Abstract
Paintings are traditionally appreciated for what is witnessed by the naked eye and the visual experience that is created from that in the centuries after the artwork is completed. However, techniques such as macro X-ray fluorescence (MA-XRF) scanning allow non-invasive methods to be used to explore below the surface and provide answers to the unanswered questions surrounding the composition, creation and history of treasured artwork.
The process of scanning artwork results in distribution maps for each of the chemical elements present and provides a mapping of their location and abundance in the painting. Paints of different colours and from different time periods will be evident on different maps depending on their chemical compositions. Producing clear and accurate elemental maps requires a combination of deconvolution, registration and mosaicking to align the maps with the correct regions on the original image.
The automation of deconvolution, registration and mosaicking through machine learning could potentially increase the speed and accuracy of obtaining the elemental maps of artwork which is still largely completed as a manual task. The goal of this project is to optimise the deconvolution of data retrieved from a variety of modalities and improve on current registration methods using machine learning techniques. Therefore, providing an easier pathway for researchers in the cultural heritage sector to expand the knowledge we have of major artwork.
The process of scanning artwork results in distribution maps for each of the chemical elements present and provides a mapping of their location and abundance in the painting. Paints of different colours and from different time periods will be evident on different maps depending on their chemical compositions. Producing clear and accurate elemental maps requires a combination of deconvolution, registration and mosaicking to align the maps with the correct regions on the original image.
The automation of deconvolution, registration and mosaicking through machine learning could potentially increase the speed and accuracy of obtaining the elemental maps of artwork which is still largely completed as a manual task. The goal of this project is to optimise the deconvolution of data retrieved from a variety of modalities and improve on current registration methods using machine learning techniques. Therefore, providing an easier pathway for researchers in the cultural heritage sector to expand the knowledge we have of major artwork.
People |
ORCID iD |