Blog

Using Deep Neural Networks to Revolutionize the Image Capture Process

Wim De Maertelaere
Wim De Maertelaere

New developments in Deep Neural Networks allow you to automate how important information is identified, captured, and/or classified. You don´t need to manually sort and index information anymore. Let the machines take over with better performance, accuracy, and consistency.

Most analytical tools rely on statistical and mathematical models to automate the capture process, but this only works well for well-defined content types without too many variations. Deep Neural Network technology takes this one step further to improve accuracy. Let me give you an example of how Deep Neural Networks work for something as variable as face recognition.

We present an image at the bottom of the network (input layer) and we see that the first hidden layers can detect very small features on the image: difference of luminosity, vertical lines, diagonals, circles. 

If we go to upper layers, we see that these layers begin to detect and categorize bigger and more complex ‘objects’ like nose, eyes and other particular signs on the visages. 

The last layers will be able to categorize the faces, taking into account all of the features and characteristics from the lowest layers.

This kind of neural network has been designed and trained by Facebook in its project DeepFace that reached a 97,5% performance on visage classification. This means a performance better than humans!

Deep Neural Networks can now be used to automate the capture and scanning process. In 2016, The Association for Information and Image Management (AIIM) asked its members about their immediate priorities for improving current capture systems, and their answers clearly reveal the automation trend.

Screen Shot 2017-06-14 at 16.21.40.pngYou can have excellent scanners to improve the scanning process, but this is of no use if you manually sort, index, and/or verify the scanned information. Deep Neural Networks allow you to automate the capture process. Here at Moonoia, we managed to have up to 70% of the process fully automated.

Interested in learning more? Join the upcoming Moonoia session at the AIIM UK Forum “Using Deep Learning to Achieve Real Straight Through Processing” or visit www.moonoia.com to learn about how AI-powered data extraction technology can transform information processing within your company.

Topics: deep learning , automation , neural networks

Written by Wim De Maertelaere

Wim De Maertelaere
Like what you see? Tell the world