Introducing Our New Patent Grant
by Amit Jnagal, on May 19, 2021 11:00:00 AM PDT
A System For Extracting Data from Complex, Unconventional Images Where Traditional OCR Solutions Fail
As the CEO of Infrrd, I am committed to continually push the boundaries of innovation so that we offer the most comprehensive and transformative solutions to our customers. In fact, we created the Infrrd Research Lab whose mission is to solve the toughest problems our customers face, as well as to ensure we are the leader in AI. From global insurance to mortgage lending to healthcare organizations, companies need powerful solutions to extract, manage and glean predictive insights from valuable customer data, leveraging the most powerful AI and machine learning-based solutions. As a result, they radically transform their business as they can better scale their operations, provide a superior customer experience, bring new efficiencies to their workforce and processes, and substantially lower costs.
Indeed, we’re seeing massive innovation in the machine learning and artificial intelligence segments, and we want to be at the forefront of that innovation. Through the Infrrd Research Lab, we are dedicated to bringing in the brightest minds from MIT, Stanford, UC Berkeley, and other leading universities to put an immense focus and investment on innovation and R&D. As part of that focus, we’re thrilled to announce that we have received our patent grant for our technology innovation regarding data extraction from complex, unconventional images where traditional optical character recognition (OCR) solutions fail.
OCR often presents what we call “background noise” in an image. You usually see this in cases where a rubber stamp is put on an existing document. So instead of trying to read the characters, we first automatically rotate the stamp so its characters are horizontally aligned. Then we pluck out each character one by one and instead of running it through OCR, match it with thousands of images of characters to read the character. The results are much better than OCR engines for noisy images.
The proposed solution focuses on stamps but the underlying concepts and the solution itself can work on similar cases like logos, banners, and non-conventional documents. Apart from the OCR bottlenecks, the solution also features a different way of classification of images where the classes are extremely similar. The entire flow has a lot of modules to workaround the bottlenecks and provides accurate results. The module uses a lot of existing Deep Learning concepts and blends them with a few different ideas to make a novel solution for this problem.
The majority of our customers have an intensely high volume of variation of data and documents, structured, semi-structured, and highly unstructured. In fact, a recent insurance customer counted over 2 million variations in data structure! To effectively manage the intake and processing with logical workflows requires a powerful solution for IDP. And unconventional images are just a portion of that data variation. We essentially extract intelligence out of all this valuable customer data and put it to work for your organization.
We are thrilled to continually innovate in this ever-growing field of AI-enabled intelligent document processing so that our customers can radically transform the way they do business with 100% accuracy.