Optical Character Recognition Comparison Analysis

Extracting data from PDFs, JPGs and PNGs is essential to almost every business. If you have experience in this area, you will understand that doing this process manually is costly, time-consuming, and can lead to inevitable human errors. This could range from reading handwritten forms, performing KYC on confidential documents, or processing invoices. These are all manual jobs that are now being done more accurately, faster and cheaper by computers.

Optical character recognition (OCR), is the answer to this costly issue. How important is OCR to business efficiency? The 4 largest tech giants, Apple, Microsoft, Amazon, and Google have created, and are constantly developing, cloud-based services with OCR capacity.

Passport Parsing

MRZ highlighted image.png

With so much choice, it is a perfect chance to draw a comparison and see which performs best based on results, speed, price, and user-friendliness. For this example, we used 4 passports of varying nationalities and resolutions. Some are scanned whilst others are pictures.

We extracted the MRZ code from a passport, circled on the image, which gives all the information about the passport holder. Below are the results of the different services:

OCR comparison table (2).png

Conclusion

We would recommend Microsoft Azure. Whilst the speed and price are comparable for all, the accuracy and ease-of-use seemed miles ahead of Amazon and Google.


AMAZON TEXTRACT

Parsing data out of passports is just one of the many cases for using an OCR, but you could also look to automate the admin of hand written forms, have invoices automatically read or translating images on the fly. However, Amazon Textract has some very interesting features with table and form detection. Moreover, it has the ability to integrate a human review system, but you must ensure that the files are of good quality otherwise you may have some accuracy issues.

NANONETS

If you have bespoke projects, Nanonets has a fascinating machine-learning platform, which allows for extraordinarily accurate models to be made. Additionally, it allows for integration with API calls or you can build upon their pre-built models to extract data from receipts, invoices and passports from day 1, in a very user friendly interface.


As always, you need a way to get the images into the cloud, call the requests, prep, and cleanse the response. Alteryx is the ideal tool for these processes. If you have data locked away in images and think we could help then please contact us for a demo on how we can get value from your specific use case.

Previous
Previous

How To Collect Your Company’s Big Data Without The Big Hassle

Next
Next

Alteryx’s Global Virtual Inspire is back: 19-21 May 2021