Mobile technology business applications make working easier and more convenient. With Ephesoft’s new Transact Mobile SDK 4.0 release, the focus has been to expand reliability, accuracy and use new capabilities to advance the product. One of the more sophisticated features is the use of Deep Learning methods, that enhance document capture for mobile.

In an interview with Dean Hough, VP of Engineering and members of his team, we found that this version applies the latest evolution in image quality, deep learning, architecture, access and security to give customers the most flexibility, accuracy and processing capture technology available to organizations. Here’s a look at some of the features and some use case examples:

  1. Auto-capture and auto-alignment of image with specialized camera SDK

Now, mobile devices can normalize, align and create the best quality images, just as good as topline scanners. For example, in mobile banking, customers can take a photo of a check and upload to their bank. If the photo is crooked or blurry in places, the image will be cleaned up. This is important to the financial services industry because not only does it normalize the image, when it is synced to Transact, customers will have full end-to-end capabilities of taking the image data, to then classify and extract the information.

  1. Live-edge detection via Deep Learning

Deep learning acts as a neural network (in the human brain) for Ephesoft’s mobile live-edge detection technology. The code is written to emulate our eyes, colors and patterns to detect edges of documents or images. Deep learning is a much more insightful detection because of its predictive capabilities, drawing on how humans think and learn. Deep learning uses open source image processing and deep learning libraries to create a state of the art live-edge detection system. Credit card recognition is a new feature that utilizes deep learning to not only detect edges, but extracts text and numbers.

This system has been previously trained and therefore, can predict edges, even when the camera is on smooth or noisy (textured) surfaces because it is already part of a sophisticated platform. For example, when you compare an image of a credit card on different surfaces, the deep learning technology easily shows the edges of the card – and predicts its edges if it can’t “see” it. Deep learning takes capture to the next level.

  1. Continuous-capture and re-capture

Continuous-capture and re-capture take continuous photos, so it’s similar to a video stream. The system then selects the clearest frame. This is most useful when users are only capturing several documents, not hundreds or thousands.

  1. Image quality

The SDK provides tools to gain the best images to use for OCR and pattern capture. Users can customize color filters, contrast, lighting, black and white, and grayscale modes to gain the best quality images. Binary (or black and white) and grayscale images yield the best results for OCR while RGB colors are better to detect patterns. Better image quality leads to better classification, document validation and throughput.

  1. Privacy and security enhancement for captured images

In a world of frequent data breaches, privacy and security are essential. Ephesoft has created a private directory or gallery where images are stored when you are using your mobile device. When you exit the application, the directory is erased. Similarly, the SDK supports run-time user permission for storage and camera access. Other applications (like Google Images or Facebook) won’t look to access sensitive documents because they are secured, or the private directory files are erased. The system asks the user if capture access is allowed.

  1. Support for ARM 64-bit architecture

The architecture supports more memory and meets new hardware capabilities. Not only does this allow the mobile technology to sync with other modern 64-bit technology, but because there is more memory available, users and companies can work more efficiently and process faster.

  1. Online and offline batch processing

Customers can now complete OCR, classification and extraction processing on a mobile device offline. Once the device is connected to the Transact server, it will automatically update batch classes or any new information. In other words, a user can fully process batches, which will be queued when a device is back online. This is helpful for field reps who might not have access to their company server. They can process data in the field – perhaps data in a factory warehouse, outside lab samples, or tank storage reports offsite, allowing for maximum flexibility.

  1. Barcode scanning

The mobile device quickly scans barcode data and extracts corresponding metadata. It can easily identify numbers that correspond to a patient identification or a product code.

  1. Credit card recognition

Credit card recognition is a new feature that utilizes deep learning to not only detect edges, but extracts text and numbers. Historically, credit card recognition has been challenging, so this specific functionality will help users process faster.

The combined feature set will drastically improve capture and in turn, results in efficiency, reliability and accuracy of data. If you are ready for a demo or are considering expanding your Transact to include mobile capabilities, please contact us at

Watch the demo