Cropilot: AI-powered image processing

Saving cultural heritage is being slowed down by small digitisation capacities

When digitizing books and other documents, cropping scans is one of the most repetitive and time-consuming steps. In many memory institutions, this step is still done manually or using outdated tools.

Moreover, the cropping process itself does not require professional decision-making at all — yet it takes up a significant part of the librarian's working day. The result is limited digitisation capacity, fluctuating output quality and unnecessary burdening of skilled workers with routine activities.

Skilled workers don't have to do a job that a machine can handle

Automated cropping saves time and nerves - the manual, repetitive steps are taken over by AI, acting as a helpful and, above all, fast assistant that never gets tired. Our solution allows you to:

• Customize cropping parameters directly to suit a specific collection/institution
• Minimize the need for manual inspection (full automation including page rotation/resolution)
• Increase digitization capacity without personnel/cost growth through process automation
• Help ensure stable and consistent output quality
• Continuous improvement based on feedback

Key Features

Operating modes

automatic, semi-automatic, training

Compatibility with digitization lines

Direct integration into NDK lines and connection to the system ProArc. Can also be used as separate solution.

Automatic viewport detection

Detection of page viewport, automatic page rotation, and left/right side resolution.

Trimming types

Full support for internal and external document cropping.

AI vision models

The system uses pre-trained vision models for full automation of clipping.

Administration and administration

Easy access rights management and document management.

Automation

Elimination of repetitive tasks thanks to the takeover of the technical routine by artificial intelligence.

Training mode or tailor-made cuts

Each institution works with other types of documents. With ordinary documents (books, newspapers), our model will playfully cope in automatic mode, and additional training is not required. However, it's not that easy to digitize with every artifact — anything from poor input document state to atypical formats can complicate the situation. In such cases, we can train the AI model so that it fits directly to the specific digitalization practice and institution.

The learning process can be simplified as the process of training a new professional:

The newcomer first independently processes the first set of pages. An experienced colleague will check his work, correct inaccuracies and explain where the error occurred. Thanks to this feedback, the novice is rapidly improving, and he already handles the next batch independently and without previous mistakes. Our model works on the same principle.

It is possible to repeat this cycle until the desired quality and trim parameters are achieved. Thus, the model is continuously improved on the basis of feedback and after training it is able to solve even the most complex types of documents in fully automatic mode.

To put it simply: the first hundred pages will pass the expert's review, the next one will be able to process the model independently with minimal intervention and error rate. Thanks to this, an experienced worker can engage in more professional work.

AI as both a helper and a tool

The work of a librarian never ends. In order to save and preserve as many documents as possible for future generations, the digitization process must not be hampered by technical routine. Leave the clippings to the machines and devote your professional capacity where it is irreplaceable: to the content itself and its protection.

‍