|
BACK TO
INDEX >>
TECHNOLOGY OVERVIEWS - OCR/ICR/INTELLIGENT RECOGNITION
OCR & ICR - IMT Magazine
(Issue 2 - 06)
By: Parascript
OCR Technology
Optical Character Recognition (OCR) – used extensively
throughout business and government – examines scanned bitmap images of
machine-printed text and translates the characters into ASCII text files
that can be edited. For instance, paper checks contain number series written
in machine print designed to minimize recognition errors. These codes
contain bank routing numbers, the holder’s account numbers and other
information required to process paper transactions. Machine print conversion
is largely a solved problem in this application, as OCR was included in the
first commercial systems that automated machine print text recognition.
Principles of OCR Technology
Optical Character Recognition (OCR) systems
recognize only machine print. Using pattern-matching technology, OCR
translates the shapes and patterns of machine-made characters into
corresponding computer codes. Though most advanced systems are able to
recognize multiple fonts, they can process only standard fonts such as Times
Roman and Arial. Once all characters in a given word are recognized, the
word is compared against a vocabulary of potential answers for the final
result.
Character recognition then segments lines of
text or words into separate characters that are recognized by the makeup of
their component shapes. Machine-printed letters are evenly spaced across,
and up-and-down, a given page, allowing the OCR system to read the text one
character at a time. Segmentation into single characters represents a
critical recognition failure point for forms processing organizations,
because OCR recognition technology requires high-quality images with
excellent contrast, character and clarity. Any text that is less than
perfect will cause even the most sophisticated OCR systems to return
significant reductions in accuracy when processing degraded images. For
example, when characters break apart due to poor image quality, or if
multiple characters merge due to blurred or dark backgrounds between them,
recognition accuracy may be reduced by as much as 20 percent.
ICR Technology
Intelligent Character Recognition (ICR) converts
hand printed characters to their machine print (ASCII) equivalents,
representing a significant step forward in technology when compared to older
OCR systems
that only read machine print. The ability to recognize handprint
significantly broadens the range of applications that benefit from automated
ICR solutions, saving time and increasing accuracy to levels not attainable
by
OCR or human
intervention.
ICR is based on the science of neural networks
that behave like the human brain when processing information. Because ICR
can handle variations in character shape, the term 'intelligent' is combined
with 'character recognition' to describe handprint recognition.
The diversity of hand printed text does not
allow ICR systems to deliver the same overall accuracy as
OCR systems
that recognize simple machine print. However, as ICR also executes
recognition at the character level, the results are meaningful if reported
as a percentage of correct characters. ICR includes the added benefit of
developing a level of confidence in each character result, where confidence
is defined as the ability to report on itself, making a judgment about the
accuracy of its recognition. The characters that ICR considers unreliable
are sent to human operators for double-checking. ICR does not eliminate
human labor but it can reduce it.
ICR is a significant step forward compared to
OCR technology. The ability to recognize handprint characters significantly
broadens the range of applications that may benefit from an automated ICR
solution by saving time and providing accuracy no manual data entry could
ever match.
Principles of ICR
Technology
Hand printed characters are created by humans,
so understanding and interpreting the patterns of human writing is far more
complicated than converting simple machine print, because no two people ever
write identical characters. Factors such as mood, environment, or stress all
conspire to create variations in character writing, causing individuals to
form characters differently each time they write or fill out a form.
Variations will even appear within the same word, depending on where a
character appears. Also, keep in mind that hand printed characters are never
evenly spaced across the page, making it difficult for recognition systems
to reliably segment words into their component characters.
Like
OCR
engines, ICR engines execute recognition character-by-character and start by
segmenting words into their component characters. Because ICR technology
recognizes separate words or word combinations, such as form fields, letters
cannot be written sloppily or stuck together.
People read text by scanning entire words, not
individual characters. When a person is not clear about, say, whether a
character is a 'U' or a 'V', he or she makes a decision based on context
rather than the shape of the character. ICR systems (like the most advanced
OCR
systems) try to imitate this human approach. They use dictionaries that
contain possible field values, facilitating word recognition by
combining primary recognition results with alternate choices, and then
analyzing available alternatives. While ICR is more robust than
OCR in handling
human printing, as with OCR engines, dictionaries are employed after the
recognition process, not during it. Therefore, if a correct guess was not
generated during the character segmentation and recognition process,
validation with vocabulary lists does not improve the result. Clearly,
business and industry is ready for the next leap in recognition technology
that addresses the problems inherent in
OCR and ICR
technologies.
|
