Hribar Luka
Alma Mater Europaea, Evropski center Maribor, Slovenija / Alma Mater Europaea, European Centre Maribor, Slovenia
Izboljšave muzejskega in arhivskega digitaliziranega slikovnega gradiva s postopki, ki temeljijo na umetni inteligenci oz. strojnem učenju
Enhancing Museum and Archival Digitized Image Material with Methods Based on Artificial Intelligence and Machine Learning
(Moderna arhivistika 2022, 5 (1), str./pp. 22–39)
https://doi.org/10.54356/MA/2022/FJRI7860
Izvleček:
Velikokrat naletimo na potrebo po izboljšavah lastnosti muzejskega in arhivskega digitaliziranega slikovnega gradiva, kot so npr. ločljivost, ostrina, kontrast, raven šuma ali druge pomanjkljivosti. V zadnjih desetih letih so orodjem dodali postopke, ki temeljijo na umetni inteligenci (UI), in močno povečali njihovo zaznano učinkovitost. Osredotočamo se predvsem na uporabo algoritmov za povečanje ločljivosti. Rezultati praktičnega preizkusa kažejo prednosti algoritmov UI, a zaradi mehanizma delovanja nevronskih mrež prihaja tudi do artefaktov, saj UI ne razume vsebine gradiva, ko je potisnjena v skrajne meje zmožnosti, ko je obravnavano gradivo nezdružljivo z učnimi vzorci ali ko učni vzorci vsebujejo napake ali pristranskost. Arhivisti bomo morali posebno pozornost nameniti zagotavljanju pojasnjevanja uporabljenih učnih vzorcev in metod, nadzorovati njihovo kakovost ter opozarjati na pojavnost neželenih artefaktov.
Ključne besede:
izboljševanje in skaliranje slikovnega gradiva, umetna inteligenca, strojno učenje, muzejsko in arhivsko gradivo
Abstract:
Enhancing Museum and Archival Digitized Image Material with Methods Based on Artificial Intelligence and Machine Learning
We often encounter the need to improve properties of museum and archival digitized image material, e.g., resolution, sharpness, contrast, noise level, etc. or eliminate other shortcomings. In the last ten years, artificial intelligence (AI) has been added to editing tools that greatly increase their perceived effectiveness. The paper focuses mainly on the use of algorithms to increase the resolution. The results of a practical test show a noticeable advantage of AI enhanced algorithms. Due to the internal working mechanisms of neural networks, artifacts also occur when AI does not understand the content of the material, when the algorithms are pushed to their limits of capability, when the material under consideration is incompatible with the learning samples or when learning samples contain errors or bias. Archivists will need to pay special attention to providing clarifications on what learning samples and methods have been used and how their quality is controlled, and to draw attention to the possible presence of unwanted artifacts.
Key words:
image enhancement and scaling, artificial intelligence, machine learning, museum and archival material
