OCR and PC Architecture

So just how important is your PC hardware when looking to use OCR Software?  Many of the desktop products do not take advantage of multi-core CPUs, and can have laggard performance numbers when it comes to Optical Character Recognition, Intelligent Character Recognition and Optical Mark Recognition.  Currently, playing with PSI:Capture, which offers a number of OCR options, and they have single, dual and quad core enablement in their licensing.  Dual core runs about 1.7 times the speed, and quad core gives a 2.7x improvement.

OCR and the Right Settings

What DPI should be set for optimal OCR Accuracy?

So. I get this question all the time and decided it might be good to post about it.  What is the best DPI setting for Optical Character Recognition (OCR)?

I have been at clients that erroneously believe the higher the DPI, the beeter the results, and feel pain whenever I see an OCR Scanner set beyond 300 DPI, and some even at 600 DPI!!  Holy cow, how do you handle those file sizes?

The fact remains that almost all OCR engines on the market are tuned and optimized for 300DPI for optimal conversion and recognition.  Going beyond this will provide no better results, and significantly increase your file size exponentially.  Most Document Capture companies provide image processing prioer to OCR that will allow you to scan at 200 DPI, with fairly consistent results.

OCR Software and Character Correction

Optical Character Recognition and Character Correction

So what is character correction when associated with OCR?  The OCR process provides the recognition and conversion of images to text, and in this process, there can be many characters that can be misidentified throughout the conversion process.  Typically, document capture applications provide the ability to identify commonly misinterpreted characters through a table of correction mappings.  So lets say a particular zone OCR field was designated as numbers only, and the engine interpreted an "l" for a "1" (that is an l for a one).  The correction piece of the recognition engine can provide logic to the OCR process, and make sure the text is properly interpreted. This can be really important, especially in SharePoint OCR environments where you need searchable PDFs in SharePoint.

This is just one of many ways to improve accuracy, but note you will need the right kind of OCR application that allows this feature to be enabled.

Zone OCR and Accuracy within Recognition Zones

Zone OCR Accuracy

So when doing zone OCR , or Optical Character Recognition on a portion of a page, what features do I need to ensure I have the best possible accuracy.  List below:

• Utilize a document capture application that provides some type of page registration.  The problem with using zone OCR is that most engines utilize a set template of coordinates on the page, and just repeat this "zone" on each page.  If the scanner is off, or the page skewed, you can have erroneous readings.  Page registration gives the recognition engine the ability to anchor a page feature, always referencing the zone from the set coordinates of the feature.
• Utilize a scanning application that provides the ability to perform image processing on the zone prior to running Optical Character Recognition . Removing lines, deshading, despeckling can provide a cleaner zone, and thus improve overall accuracy.
• Some advanced capture applications provide the ability to filter zones based on character sets.  This allows you to interpret the characters within a zone as say, all numbers, or perhaps a date, which provides the engine a more narrower character set for the whole recognition process.  iCapture for example, not only allows character set mapping to zone ocr templates, but also provides auto-correction for the most commonly misinterpreted characters.
• Finally, and highly recommended for the highest level of accuracy, is the ability to set a character matching filter for a zone.  This technology, sometimes called ADE, provides the ability to utilize regular expressions to ensure a match, and lets you over draw the recognition area / zone and filter to your liking.

OCR Software and Image Processing

OCR Software and Image Processing

Why is image processing so important when utilizing Optical Character Recognition Software?

In order to get the highest possible accuracy with your OCR Application, the recognition process needs to have a clean image to examine.  The most important are auto-orientation, deskew and despeckle.  The Auto-orientation process examines tha page, and makes sure it is oriented correclty for the whole recognition process.  Deskew examines the page for any skewing, whcih may occur during the scan process, and "rights" the page to make sure the text is inline throughout the page.  Despeckle takes away any speckles on the page that can be falsely identified as font characters, but also can be attributed to any misreads of characters.

Older documents may require other functions, such as font improvement and deshading to insure the highest possible accuracy in the overall OCR process.

How do I pick the right OCR Software?

In the space of OCR Software, or Optical Character Recognition, it can be confusing to say the least on which option you should pick.  It really comes down to the use case, or how you will utilize the software.  Below are some great question to ask your self:

What do I need to convert with my OCR Software? 
This question is very important, and it really comes down to what you are looking to output with your software.  Do you want a word file that you can edit, or are you just looking to create a searchable PDF?  Many engines are tuned for accuracy, and will give you the best formatted output, others are built for speed.  Omni-page is an excellent engine for creating nicely formatted output, but can be rather slow due to its focus on acuracy.  A production engine, like PSI:Capture, which offers multiple OCR choices, can give you great flebility, no matter your ouput choice.

Are they pre-existing images, or ones that I will scan?  PDFs or TIFFs?
It is really important when you are choosing Optical Character Recognition Software, to make sure that you have all the functionality you require, whether you are scanning, or just processing non-searchable PDFs from a directory.  Most of the OCR Software will let you choose the file that you perform recognition on, and others will let you scan in paper for conversion.  If you are utilizing MFPs or Scanning copiers, and want to perform OCR on the scanned documents, you may want to choose a product that performs auto-import, or one that is focused on MFP Scanning.  Also, you want flexibility in the types of file you can process, and want to be able to OCR any image type:  PDF, TIFF, JPG, GIF, BMP, etc.
How fast can I do conversions?
So, some engines are built for OCR Accuracy, others built for speed in the OCR process. Most of the desktop engines, like eCopy Desktop, provide a good mix of both.  Other engines, like Glyphreader or Docustar, provide the ability to choose whether you want speed or accuracy in your OCR results.  It is always good to choose a document capture option that allows you multiple OCR engine options to perform diffferent recognition tasks.

How ddo I get the best accuracy in the OCR ouput?
All of the OCR Software mentioned within this post reuires a high quality image for the best recognition accuracy.  With that said, a high quality scanning software with image processing options will lead to the best OCR accuracy when converting from image to text.  So what does image processing have to do with OCR Software?  The cleaner the image, the better the accuracy, and if you can deskew, despeckle, deshade and sharpen text, you will get better OCR results.