On January 19th, Parascript hosted the webinar “Keys to Successful Handwriting Recognition”. In case you missed it, you can get the recap anytime on our YouTube channel:

There’s some great content in here – and if you’re new to recognition technology, or you’ve had a bad taste in the past, this is definitely worth a look! It’ll cover:

• Common myths of ICR, how ICR is different than OCR and other recognition technologies
• Important aspects in ICR technology, including noise removal, contextual analysis and orthogonal analysis
• Understanding key differences between multiple comparison, or voting technologies
• What to look for in ICR technologies

Enjoy!

Last month on this blog I recapped my trip to the AIIM Document Management Service Providers Executive Forum. As noted, I found a healthy degree of skepticism to whether or not ICR for handwriting (including cursive) was a viable forms automation tool.

In 2012 we’re setting out to address the misconceptions and set the expectations straight. ICR can be very powerful for forms automation – but it may not be for everyone.

Feel free to join us on January 19th for a live webinar, Keys to Successful Handwriting Recognition. Here we’ll shed some light on how to optimize an ICR solution for best performance, and provide some insights into the latest in ICR technology.

Here’s the signup link, hope to see you there!

https://www2.gotomeeting.com/register/698567314

Last week, I escaped Denver during a fall snowstorm to attend the AIIM Document Management Service Provider Executive Forum in Hollywood, Florida. It was great to meet so many new and interesting people. I found the sessions on healthcare and digital mailroom particularly interesting – and Dan O’Leary and Atle Skjekkeland just hit it home with their social media presentation at the end.

What I also found interesting was a really healthy dosage of skepticism towards Intelligent Character Recognition (ICR) in forms processing. A lot of people have tried it but the results were less than desirable, so there’s understandably some skepticism around it. I’ll take a swipe at the recognition industry and suggest that this is what happens when capabilities are oversold.

ICR for forms processing can be an incredibly powerful business automation tool, and can drive a lot of cost out of forms processing. However ICR is both an art and science. The key to success lies in defining context. In check and postal processing, recognition rates can be over 95% with less than a 1% error rate. That’s because when reading a check and an address, you know what the fields are supposed to be, and have standards to compare against. That general bit of knowledge enables the software to reverse engineer the right answer.

Many forms contain constrained fields as well. Applying context to these fields can dramatically increase recognition rates, and then exception rules can be built to optimize the use of manual keying. It would be difficult to believe that we can remove the human factor completely (IBM makes a good run at it with Watson) but by applying a blended ICR and human-keying practice you’ll be able to greatly reduce processing costs and increase productivity.

Pictures from the event can be found on the AIIM Facebook page: here.

Parascript is proud to be a recipient of the 2011 Longmont Area Economic Council Community Appreciation Award for our work with There With Care. There With Care is a non-profit organization supporting children and families facing critical illness in Colorado. We dedicated the Parascript 10th Annual Art Exhibition to raise funds with a care drive, silent auction, and the sale of calendars featuring the artwork.

We are honored to have received this award and to support such an important organization that provides much needed services to families going through a medical crisis. We are also very proud to showcase the amazing talent we have in our company with our annual art exhibition. Employees and their families participate with photography, paintings, pottery, drawings, etc. The exhibition also features artwork created by children in There With Care program. We invite you to check out There With Care online at www.therewithcare.org.

Paula Barton, vice president of human resources, received the award along with Paula DuPre’, executive director and founder of There With Care.

Japanese Lady and Daughter by Sofia Prizemin, age 10.

Last week I went to the Remote Deposit Capture (RDC) summit in Orlando. This was my first RDC summit – and with a declining check volume (Federal Reserve stats show a 41.5% decline from 2000-2009) – I was half expecting an event with waning interest. Was I ever wrong!

I was pleasantly surprised to see a lot of attendance and engagement both in the general and breakout sessions. The topics were relevant, and there’s a lot of interest around how banks and credit unions can leverage the benefits of mobile capture in order to receive deposits from consumer cell phones. We learned that not only is the technology a significant convenience for consumers (and cost reduction for financial institutions) – it may be the single largest contributor to customer retention in the market today.

There was also a lot of discussion around RDC as a remote payment hub – broadening the concept well beyond checks – as well as its application in lockbox environments.

Many of Parascript’s partners were present as well – providing the enabling technologies behind RDC. It was great to connect with them and learn more about how our technology continues to serve this market.

Hope to see you there next time!

The U.S. economy and the USPS are intrinsically tied

There’s a lot of speculation about what will happen with the U.S. Postal Service (USPS) when it officially runs out of money at the end of September. And the pundits are doubling down. Political brinkmanship will likely run until the last minute, perhaps even a little bit beyond. But the bottom line is that it is inconceivable that the USPS would not get relief from the government.

The USPS moves too much of the country’s economy to let it suspend operations. A shutdown of even a few days would be a financial paralytic to an already ailing economy. From packages to bills to payments, the US economy cannot function without the USPS.

For these reasons, relief is likely. Yet despite the fact that the USPS has become a relatively efficient operation (Parascript technology plays a significant enabling role), it still needs to deal with declining revenue, a massive expense load and a business model that is not easily adapted to the times. Cuts will have to be made. Restructuring is not a matter of if, but what and when.

If your business is co-dependent on the availability of the USPS (in some way or another most businesses are); you shouldn’t lose sleep. This will get worked out, although some changes — such as the loss of Saturday delivery — may present near-term challenges. If you’re personally caught up in or affected by the political process; you have our sympathies as the next several weeks are likely to be stressful times.

While the industry historically has benchmarked recognition products based upon character-reading accuracy, in reality the studies may not be an accurate indication of attainable performance. New recognition products take a different approach, focusing on recognition of the entire word in addition to individual characters.

A sophisticated technological approach delivers real improvements in overall recognition rates, resulting in cost savings for companies who can implement this approach. In the following pages, you will receive a primer on the technological evolution in character recognition and how advanced recognition engines combine all three reading methods – OCR, ICR and NHR – to deliver maximum accuracy.

Evolution of OCR and ICR

The two most common types of traditional recognition engines include optical character recognition (OCR), which is used for recognizing machine print, and intelligent character recognition (ICR), which is typically used for recognizing handprint and cursive.

In the 1950s, the first OCR device, dubbed “Gismo” by its creator, could interpret machine print, Morse code characters, and musical symbols. The Reader’s Digest Company used this first OCR system to convert typewritten subscriber information into computer punch cards.

About 30 years later, Kurzweil Corporation introduced a system that also accommodated handprint – the first ICR machine. The company’s founder, Ray Kurzweil, became known as “the father of the reading machine” with a device that converted machine-printed and Internet-based text into synthesized speech.

Both OCR and ICR are based upon analytical approaches that consider words as sequences of characters rather than whole words and phrases. Based upon an analysis of the lines and curves, OCR and ICR technology make a best “guess” at each individual letter, number or symbol.

Once the string of “guessed” characters has been determined, a look-up table is used to most closely associate or match the string of characters. This is a “static” process in that the comparison occurs after the recognition takes place. In cases where there’s an absence or an unusual placement of elements the performance levels will be lower. And, while look-up tables are usually context sensitive to improve accuracy levels, the process is slowed when comparing all the “guessed” alternatives.

Handwritten Character versus Word Recognition

To overcome the limitations of OCR and ICR systems – in order to recognize unconstrained text – new technology had to adopt an essentially “human” method of reading and recognizing. The system needed to take into account the entire word or phrase in addition to its individual characters. Just imagine the difficulties involved in developing a technology that recognizes all handwritten text, where each person’s handwriting style is so unique. The variations in the way people make letters and numbers are almost infinite. How could a technology be developed that could read a field from a document and provide a recognition result based on the interpretation of an entire word or phrase?

Natural Handwriting Recognition (NHR)

The concept of natural handwriting recognition was first introduced in 1976 in a scholarly article by noted Russian scientist Shelya Guberman. Over the next 20 years, scientists worked to develop handwriting recognition and bring the technology to a market ready state. During that time, the group formed a joint Russian/American venture that evolved into the company now known as Parascript.

Natural Handwriting Recognition (NHR) technology takes a “holistic” approach to recognition in that the fields on a form are processed as complete words. The recognition process involves a combination of highly advanced mathematical algorithms, neural network applications, image interpretation using a special description language, and context validation that happens during the recognition process.

Defining OCR, ICR and NHR: Exploring the technology and methods that are driving today’s biggest character recognition efficiency gains in the commercial arena

In the next few posts, we’ll share the basics and evolution of image and character recognition; discuss the three major types of automated image, optical and intelligent character recognition; and explore the powerful performance and productivity gains that can result from combining the best of all three.

The History and Driving Forces of Automated Image and Character Recognition

Herman Hollerith introduced punch-card automation at the time of the 1890 U.S. census. The world’s first data entry operators manually encoded handwritten ledger entries from holes in roughly 50 million punch cards. Processed by hand, these early cards defined an American population of close to 62 million.

Over the next 60 years, manual data entry represented the only available method for creating machine-readable information. Manual data entry is still the most versatile of all recognition techniques because it uses human brainpower. Data entry operators can overcome almost all normal irregularities, including characters that are either stylized or misaligned to the form’s boundaries.

However, the people- and time-intensive elements of manual data entry are its downsides. The process often takes too long and is easily the most expensive way to digitize data. As such, it may be cost prohibitive given today’s massive amounts of documents that require conversion to digital data.

Machines that Read Like Humans

For these reasons, there is an increasing need to reproduce the power of the human brain in a mechanized process where handwritten, printed and imaged documents can be interpreted and converted into data records with a high level of accuracy – accommodating all of their irregularities with a minimal amount of human intervention and able to be processed in-house or through outsourcing. This highly efficient process is the ultimate prize in the story of commercial character and image recognition.

United Business Mail is a presort mail services company providing postage savings to businesses for first class letters and flats, standard/non-profit letters and flats, bound printed matter, presort parcels, and more. Additionally, United Business Mail provides complete mailroom and metering services to allow customers to reduce the overhead costs associated with these functions.

Annually, United Business Mail processes over 250 million pieces of first-class and standard mail from hundreds of customers across the country.

United Business Mail had used Parascript, as the only optical character recognition (OCR) technology to process handwritten mail, at its Minneapolis facility for nearly 10 years and its Phoenix facility for over five. When it opened its Los Angeles facility, management of the mailing company determined it only made sense to invest in the best handwriting recognition technology for letters and flats.

United Business Mail has achieved significant gains on finalizing handwritten addresses with Parascript’s OCR technology on its BÖWE BELL + HOWELL PTI High-Speed Multi-Line Optical Character Reader (MLOCR) at its centers in Minnesota and Arizona. Thanks to the solution, United Business Mail is able to obtain automation rate discount and improve deliverability of the mail by applying a barcode before entering the mail into the USPS plant for continued processing. Additionally, Parascript’s click, with pay-as-you-go, model significantly reduced the cost of the initial investment.

To find out more about how United Business Mail leveraged Parascript, please download the free case study here.

In this post, we explore the cost of errors and rejects in a hypothetical company’s manual processing situation.

With recognition, there are two important variables affecting success rates that are specific to a customer: errors and rejects. Errors are issues involved with false data recognition. In this instance, data is passed onto another system that is incorrect. A reject occurs when the confidence rate of a particular recognition process does not meet a specific threshold and requires manual intervention. Both incur a cost, but both costs can be quite different.

Both incur a cost, but both costs can be quite different. How so? Errors involve sending incorrect data to another system. Costs here could be trivial. Let’s say the error is associated with a name used for marketing correspondence. The result is that the person receives correspondence but with an inaccurate name. Or it could be large: the error could be associated with a SSN which is used for some financial deal.

For rejects, the cost is typically associated with the effort required to review the data and correct it manually. If the business is outsourcing this effort to a low-cost area, then it might be trivial. The key here is to balance the cost of manual interaction with the cost of bad data going into the system.

A Hypothetical Example – the Cost of Manual Processing

The context around the sample could be as follows (numbers are hypothetical and don’t represent a specific businesses’ actual costs): the company currently does manual processing of claim forms with a manual keying error rate of around 2% of all data entered. The cost of manual data entry is $5 per form and they process 100,000 forms per year. Thus the total manual cost is $500,000 annually. Then add the cost of errors. The cost of 2% errors contributes to additional costs of $200,000 (let’s say if everything was erroneous, it would cost the company $10m). Thus, the total cost of manual processing is $700,000 per year.