Case Study

Neurotechnology developed an image-to-text solution for the Lithuanian Audiosensory Library (Lietuvos Audiosensorinė Biblioteka - LAB) to improve the accessibility of visual content in digital publications for people with reading difficulties.

The same workflow turns illustrations, diagrams, maps and artwork into screen-reader-ready descriptions – in Lithuanian, and adaptable to other languages.

The LAB sought a way to make visual information in digital publications more accessible to its users. While textual content can already be processed by assistive technologies, visual elements often remain inaccessible, limiting access to educational, scientific and cultural materials.

According to LAB, more than 300,000 people in Lithuania are unable to read standard printed text due to various disabilities; only around 20 percent of books published in the country each year are available in accessible formats.

The solution was developed as part of LAB's "ELVIS – Adapted Media Platform: Expanding Access to Inclusive Cultural Content for Users with Individual Needs" project. It is being integrated into ELVIS, LAB's virtual library for accessible publications, where it will support the preparation and delivery of accessible content.

Selected through LAB's public procurement process, Neurotechnology's solution is planned to be used at least until 2031. By automating parts of the image description process, it can reduce manual work, accelerate the preparation of accessible publications and help maintain consistent quality.

Before launching the public procurement process, LAB conducted a detailed market analysis of available tools for generating illustration descriptions.

"Our analysis revealed clear differences between the available solutions. Not all tools could recognize and extract text from illustrations, while Neurotechnology's solution performed these tasks very successfully. We also evaluated the quality and accuracy of generated descriptions and their suitability for accessibility requirements." Erika Buivydienė
Deputy Director of the Lithuanian Audiosensory Library and the manager of the project

The solution uses a combination of computer vision, optical character recognition (OCR) and large language model technologies. Computer vision algorithms identify visual elements within an image, while OCR extracts embedded text. This information is then processed by language models that generate detailed descriptions in the Lithuanian language.

By combining image understanding and natural language generation in a single workflow, the system can automatically transform illustrations and various image types into accessible text suitable for screen reader applications.

The technology can also assist readers with dyslexia by recognizing and extracting text contained within visual elements, including titles and quotations.

The system is designed to process various types of visual material, including:

1. Illustrations – The tool identifies key objects, people, scenes and visual elements, then converts them into a structured text description.
2. Diagrams and schemes – The AI solution can interpret visual structures and describe the main components of diagrams or schematic images.
3. Maps and historical visuals – Complex visual materials, such as historical maps or educational illustrations, can be described in a way that helps users understand the essential information.
4. Artwork and cultural content – Detailed descriptions may include the type of image, dominant colors, emotional tone or artistic style, depending on the content.

"Artificial intelligence can create real social value by making visual content understandable through clear textual descriptions. This can significantly improve access to culture and information for people who face barriers when reading printed publications." Vytas Mulevičius
NLP Team Lead at Neurotechnology

As a result, visual information that would previously require manual adaptation can now be automatically transformed into accessible Lithuanian-language descriptions, helping expand access to educational, scientific and cultural content.

Because the solution produces clean, structured text, the same descriptions can also be voiced with text-to-speech to create audio versions of visual content – extending the workflow toward audiobook-style production for users who prefer listening to reading.

LAB specialists who are already using the Neurotechnology solution highlight its ability to use contextual information when generating illustration descriptions as one of its strengths.

"This helps ensure that the AI is used effectively and appropriately. Generating descriptions directly in Lithuanian also makes the process easier. There are still occasional grammatical errors or unsuitable word choices, but the process of creating descriptions is already becoming simpler." Justinas Kilčiauskas
IT specialist and accessibility expert at LAB

The Lithuanian Audiosensory Library project demonstrates how artificial intelligence can be used to improve information accessibility and create more inclusive digital experiences.

The project contributes to broader efforts in Lithuania and across Europe to make digital services and electronic publications accessible to all users. Since June 28, 2025, e-books in Lithuania have been subject to legal accessibility requirements.

Although developed for Lithuanian, the same pipeline can generate descriptions in other languages, making it suitable for publishers and educational content providers working across multiple markets.

For public-sector bodies, libraries and publishers, automated image descriptions also help meet growing digital-accessibility requirements such as the European Accessibility Act and WCAG.

Neurotechnology develops custom artificial intelligence solutions based on natural language processing, computer vision and other AI technologies for public and private sector organizations, helping automate complex information processing tasks and improve access to digital content.

The ELVIS project is funded under the "Next Generation Lithuania" Economic Recovery and Resilience Plan.