Optical and intelligent character recognition - OCR and ICR for friends - recognise printed text and handwritten, freestanding scribbles respectively. In itself nothing new under the sun, because both have existed for decades. But the evolutions in machine learning have re-drawn the whole picture in recent years.
State of the art OCR and ICR systems are now trained from neural networks: they receive large amounts of data and learn, for example, to 'read' by swallowing thousands of copies of the letters A, B to Z and learning what makes such a letter specific and what variants of it all exist.
The limits of the possibilities of OCR and ICR are constantly shifting. Where a scan used to have to be nice and straight and in high contrast, the rules are now much more flexible. A wrinkled or angled sheet of paper is no longer a problem, just like pale text on a dark background or print in combination with handwriting
We closely follow various systems in this field, both cloud-based and SDK installations. To get even better results, we like to combine multiple OCR systems, so the chance of recognition is much higher. The result is reliable systems that work independently of any software provider but read your documents in no time at all.