Scientists are developing a neural network platform called "Digital Ark" that will be able to recognize details even where a specialist sees only a dark spot, and add missing fragments in accordance with iconographic canons. This was reported to "First Technical" by the press service of RTU MIREA.
An icon is a complex language of symbols, attributes, and characters. The neural network must learn not only to analyze the image, but also to understand the meanings: who is depicted, what plot is embedded, to which school the writing belongs. For this, scientists have already collected a dataset of approximately 6,000 icons.
We conducted an analysis and realized: thousands of icons today are either not identified or are being destroyed, and manual restoration is a long process and not available to everyone. Our "Digital Ark" is an attempt to save the memory that is irretrievably disappearing.
The first site for the project was the Nikolo-Solbinsky Monastery in the Yaroslavl region. Students have already begun photographing icons to train the neural network. Negotiations are underway with the Solovetsky Monastery and specialized universities.
Read more on the topic:
- AI will raise archives: Novosibirsk residents will be helped to find relatives in old photos
- Neural networks lie convincingly: a dangerous pattern revealed by a study at Perm Polytechnic University
- AI will not write a new Tchaikovsky symphony: IMEMO RAS explained why art remains the domain of humans