In the ever-evolving era of digital payments, bank cheques have managed to stand their ground, widely recognised and accepted. Information retrieval from photos of cheques exhibit a plethora of problems such as no standardization across banks, convoluted background imagery and pose/illumination challenges.
Now extract all relevant fields from mobile quality photos of bank cheques across banks in India.
Getting creative with menus is easy, transcribing them is hard. Did we say bloody hard ? In the wake of an infinite structural possibilities, difficult to recognize and read fonts, creative imagery in the background, the problem of retrieving structured information from the menu is a convoluted one.
Extract structured information from unsctructured menus and integrate directly with your catalogue management systems.
Transcribing documents in the form of applications, forms, etc. is central to a lot of industries such as Banking, Finance and Insurance, Healthcare amongst others. We are committed to solving challenging challenging problems involving information retrieval from documents at scale.
Do you have core processes which depend on digitization of documents at scale ? Contact us to see how Mjölnir can help your business.
We, as both, consumers and enterprises live in a fools world with the prepetual hope that sooner or later everything will be digitised and paper would become obsolete. However, poor digital adoption rates and the difficulty for some demographics to move off paper have made businesses realise that the paper problem will remain.
Traditionally the processing of transcribing information from paper into a digital format is handled by a large workforce of onshore manual data entry workers or by outsourcing it to offshore BPOs. Unfortunately, since back-office service workflows don’t contribute directly to revenue, they lack visibility and prioritisation.
The question that now arises is can one INNOVATE ?
With Merak’s state-of-the-art information retrieval service, enterprises can now make paper move at the speed of digital without having to either rip in-place processes or completely move to digital. Based on the all mighty mythical hammer of Thor from the Norse mythology, Mjölnir is powered by deep learning models that detect and recognise text in any natural image.
Mjölnir can be tailored to suit every clients' specific needs to help the retrieve structured information from photos of documents. Combining principles from machine learning, image processing and other advanced algorithms, Mjölnir identifies what is important to you and spits the same in a format that can be consumed downstream by your systems in-place.
Say goodbye to your transcribing woes with Mjölnir and make paper move at the speed of digital ...
Eliminating/assisting manual entry workflows cuts down cost by a staggering 75% and increase straight through processing.
Build proprietary datasets from rich historical archives to improve operational efficiency and enable real-time business intelligence.
Reimagine data automation without having to rip or replace current systems. Exceed human data entry accuracy of 85%-95% and enable paper to move at the speed of digital.
The short and simple answer to this question is NO. The longer version of the same revolves around the technology used in both practices and their performance on real-world images.
The optical character recognition problem can be broken down into two parts, text detection and text recognition. General purpose OCR engines do not focus on solving the first part well, as a result of which text detected from an image may be inaccurate and incomplete. Traditional OCR engines come with a tonne of baggage in contrast to current deep learning based methods which are extremely pliable. They work best when there is clean black text on solid white background in a common font, text is approximately horizontal and the text height is at least 20 pixels.
With a clear edge, deep learning based methods are able to out-class traditional OCR engines by a margin of more than 100% on real-world images. Studies suggest accuracies of these methods to range from 90% to 99% in contrast to only 40% to 60% as demonstrated by OCR engines.