Deep Learning API for correcting rotation of documents

We work extensively for financial institutions for processing documents using our Deep Learning based solutions. These solutions are mostly in the form of pipelines with various modules such as classification, document rotation, OCR, text extraction, confidence score etc. Checkout this blog for one of such pipelines.

Almost every-time we put in a check to correct for rotation of the document in our pipelines - that means ensuring that the text is right way up. This ensure that the next steps of the pipeline like OCR and text extraction get the documents in the correct orientation.

The reason for erroneous rotations is that either:

• The documents are clicked through mobile phones and camera can be turned either ways while clicking the image
• A single PDF document would have different pages scanned in at different rotations
• Some times images have erroneous 'exif' information in their images which force changes the image rotation randomly

We are now releasing API for correcting rotation of the documents. API takes a document image URL as an input and returns - class of the document rotation i.e. the document can be either:

• Up
• Down
• Clockwise
• Anti Clockwise

Though there are other mechanisms using image processing libraries to correct for rotation we trained our own Deep Learning Models for better accuracy and easy train-ability. Them model is an ensemble of Deep Learning models hosted on cloud GPU platform and accessible through API calls.

This has been trained on 3 Million documents of all kinds - Forms, KYC documents, Bank Statements, Annual Reports, Certificates, Receipts etc. In most of the practical scenarios the accuracy is 99.999% i.e. there is one error in predicting correct rotations in over 100,000 document image queries. This API is being used on a daily basis on thousands of document as part of many pipelines and is robust.

We are soft launching the API today. If you are interested in testing the API, just drop an email to my colleague aman.verma@silversparro.com for the documentation and API endpoints. First 1000 API calls shall be absolutely free.

Silversparro Technologies aims to help large enterprises solve their key business problems using expertise in Machine Learning and Deep Learning. Silversparro is working with clients across the world for computer vision, Natural Language and Large Transaction data use cases working for manufacturing, BFSI, healthcare verticals etc. Silversparro is backed by NVIDIA and marquee investors such as Anand Chandrashekaran (Facebook),Dinesh Agarwal (Indiamart), Rajesh Sawheny (Innerchef) etc.

Silversparro is founded by IIT Delhi Alumni – Abhinav Kumar Gupta, Ankit Agarwal and Ravikant Bhargava and is working for clients such as Viacom18,Policybazaar, Aditya Birla Finance Limited, UHV Technologies etc.