Adobe is getting into the mobile scanning game with a new free iOS and Android app aimed at providing users with high-quality images of physical content they want to capture digitally.
It’s called Adobe Scan, and it works similarly to a bunch of other apps already available. Users point their smartphone cameras at whatever document, white board, or presentation screen they want to capture, and the app automatically crops the image to just pick out a document.
The clearest difference between Scan and other apps, like Microsoft Lens, is that the app integrates with Adobe Document Cloud and automatically performs optical character recognition on the PDFs it generates. Users can then copy and paste text from those documents into other files.
But Adobe also put in a lot of work on building machine intelligence into Scan from the ground up, according to Akhil Chugh, a senior product manager for Document Cloud. The app uses a variety of machine learning and image processing algorithms to help deal with the myriad issues that arise with mobile scanning, like trying to differentiate a document from a similarly colored background and figuring out which areas of a document warrant optical character recognition.
“Our goal has been to create a digital document that’s as good as you would get from a flatbed scanner, or as good as the real document,” Chugh said.
Specifically, Adobe is using genetic algorithms (so named because they’re generated by simulating natural selection) to handle document boundary detection. For differentiating text from images, Adobe is using tree- and logistic regression-based classifiers. Those machine learning tools are a part of Sensei, Adobe’s name for its internal ML framework and features.
Adobe is counting on its document expertise and machine learning capabilities to boost Scan in a crowded market. The app is free for people to use, but it requires the creation of a Document Cloud account. Scanned files are stored in Document Cloud automatically but can also be saved to other services.
While it’s possible for Scan to capture images without a network connection, all of the OCR work is handled in Adobe’s cloud. That means text recognition requires sending images of the document over the internet, which may not be an appealing option for some sensitive content.
The launch of Scan is a significant milestone, but Adobe plans to further refine the app over time as part of its Document Cloud portfolio. The company has invested a great deal in handwriting and font recognition, for example, so it’s possible features like that could show up in Scan going forward.
On the machine learning front, the team is investigating how to best implement deep learning in Scan and across Document Cloud. It’s also looking into using generative adversarial networks, which are designed to help create content.