Powerful OCR Annotation Tool

An OCR annotation tool like Kili Technology helps you accelerate and smoothen the labeling process of documents that include text, such as PDFs or images.

You can annotate a pdf in different ways: place bounding boxes around images and write the text transcription (OCR annotation), place masks around images (object detection), highlighting text passages (named entity recognition), classify pdf at document level or page level.

Today you can expect more than 90% accuracy for OCR based data extraction using deep learning and other ML technologies - even for hand written documents. The trick is not to rely on OCR technology alone for reliability. What works best is the combination of machine learning technologies and multiple OCR engines. Different OCR engines have different strengths - some work really great on scanned documents others are good at images captured from mobile. But once you deploy data science and machine learning technologies on top of the extracted data, you land up with something that is way more potent than vanilla OCR. This is particularly useful for handwriting where the quality of writing varies a lot. Deep learning can also be used to identify relevant sections for extraction and well as classifying documents before extraction. This helps you understand what to expect from extraction.

A convolutional neural network would be the right type. The specific architecture would be hard to specify as it depends heavily on the dataset, but if it is a low resolution image we would recommend fractional max pooling+Bottlenecked DenseNets, if it is a high resolution image we would recommend Inception V4 or Inception ResNet-v2. Anyway we encourage you to experiment and see what works best.

We would suggest Optical Character Recognition: An Illustrated Guide to the Frontier offers a perspective on the performance of current OCR systems by illustrating and explaining actual OCR errors. The pictures and analysis provide insight into the strengths and weaknesses of current OCR systems, and a road map to future progress.

Abbyy FineReader, OmniPage Ultimate and Adobe Acrobat Pro. When it comes to open source: Tesseract.

Our customers have many different use cases for OCR annotation, and some of them include detecting serial numbers and information from invoices or blurring confidential information from images with text.