ERA: A new, fast, machine learning-based software to document rock paintings.
The present study proposes a new software program to help researchers identify rock paintings from digital images, rapidly producing high-quality documentation, in a user-friendly way. The three RGB colour channels of the digital image are first decorrelated and then stretched, a well-known techniqu...
| Publicado en: | Journal of Cultural Heritage Vol. 58; pp. 91 - 102 |
|---|---|
| Autores principales: | , , , , , , , |
| Formato: | Artículo |
| Publicado: |
Elsevier B.V.
Nov2022
|
| Materias: | |
| Acceso en línea: | Ver este registro en EBSCOhost |
| fields | @attributes: recordID: 1 pdfLink: plink: https://search.ebscohost.com/login.aspx?direct=true&db=hlh&AN=160368506&site=ehost-live header: @attributes: shortDbName: hlh uiTerm: 160368506 longDbName: Humanities International Complete uiTag: AN controlInfo: bkinfo: jinfo: jid: 12962074 KK8 jtl: Journal of Cultural Heritage issn: 12962074 maglogo: N pubinfo: dt: Nov2022 vid: 58 pid: 467 pub: Elsevier B.V. artinfo: ui: 160368506 10.1016/j.culher.2022.09.018 ppf: 91 ppct: 11 formats: tig: atl: ERA: A new, fast, machine learning-based software to document rock paintings. aug: au: Monna, Fabrice Rolland, Tanguy Magail, Jérôme Esin, Yury Bohard, Benjamin Allard, Anne-Caroline Wilczek, Josef Chateau-Smith, Carmela affil: ARTEHIS, UMR CNRS 6298, Université de Bourgogne–Franche Comté, 6 Boulevard Gabriel, Bat. Gabriel, Dijon 21000, France Musée D'anthropologie Préhistorique de Monaco, 56, boulevard du Jardin exotique, MC 98000, Monaco Archaeological Research Center of the National University of Mongolia, n°202, 2nd Building, 1 Ikh Surguuli Str., Baga Toiruu, Ulaanbaatar 14200, Mongolia Cadoles, 29 bis rue de l'Arquebuse, Dijon 21000, France Centre André Chastel, UMR CNRS 8150, Faculté des Lettres de Sorbonne Université, 2, rue Vivienne, Paris 75002, France Department of Archaeology, University of Hradec Králové, Rokitanského 62, Hradec Králové 50003, Czech Republic CPTC, EA4178, Université de Bourgogne, 4, boulevard Gabriel, Dijon 21000, France su: Rock paintings Supervised learning Independent component analysis Machine learning Support vector machines Inpainting sug: subj: Rock paintings Supervised learning Independent component analysis Machine learning Support vector machines Inpainting keyword: Colour channel Colour space Open source Whitening transformation ab: The present study proposes a new software program to help researchers identify rock paintings from digital images, rapidly producing high-quality documentation, in a user-friendly way. The three RGB colour channels of the digital image are first decorrelated and then stretched, a well-known technique used by remote-sensing specialists for over thirty years. In contrast with the approaches previously developed specifically for rock art, several data-whitening algorithms are used at this step: (regular) principal component analysis, zero-phase component analysis, Cholesky decomposition, and independent component analysis. These transformations produce different arrangements of the colour information, which nevertheless share some important properties (e.g. the covariance matrix of the new channels equals the identity matrix). The decorrelated data, previously stretched and scaled to fit the RGB space, are then converted into various colour spaces (selected from among the most popular): XYZ, HLS, HSV, LAB (CIELAB), Luv, CMY(K), YCrCb, and YUV. The most subtle colour variations will be better perceived in some of these newly produced, contrasted, false-coloured images. The researcher can then take advantage of supervised machine learning algorithms to isolate painted figures. At this step, binary pixel classification is performed either by logistic regression, support vector machine, or k -nearest neighbours, possibly including confident learning. There is no need for complex tuning at any point during the procedure, which lasts a few minutes at most, while a posteriori cleaning of the produced document is minimal. The software, written in Python, is provided as a stand-alone executable program for Windows, for broader diffusion, and as open-source code, which can therefore be adapted to the evolving needs of the community. [Display omitted] pubtype: Academic Journal doctype: Article src: R language: English refInfo: copyright: @attributes: flag: Y dt: @attributes: year: 2022 holdings: @attributes: islocal: N |
|---|