Cropping and straightening is a tedious, time consuming, and error prone task. In many heritage digitization workflows a third or more of total production time is manual cropping and straightening. AI Crop aims to drastically reduce the time users spend cropping and straightening.

Legacy automatic cropping tools (e.g. AutoCrop in Capture One CH) use computer vision (i.e. shape finding) that struggles with low-contrast, soft edges, unusually shaped materials, distracting elements in the frame (e.g. a bone pick or hand in the corner), or material tool close to a target or machine element.

AI Crop in DT Nexus is not using legacy shape finding technology. Instead, it uses an artificial intelligence model trained on tens of thousands of real-world images from partner institutions such as the US Library of Congress, University of Western Australia, the National Library of Sweden, and the Swiss National Library who generously allowed us to train our model on their collections. We diligently labeled the components in these images, such as pages, fanning, gutters, partially visible covers, and targets.

The AI model underpinning AI Crop understands the material inside of the frame. It's not seeing a rectangle; it's seeing a page, or a cover, or a target. This makes it far more resilient and all around "smarter" than legacy cropping methods.

When you use AI Crop to crop an image, the processing happens entirely locally. The model, the image, the processing, and the resulting crop reside only on your computer. Your images never leave your computer and AI Crop does not use them to train or refine the model.