Thursday, December 27, 2012

Understanding Documents

Like every company, Athento has a mission, sort of a reason to exist.
"Understanding Documents" would be the match for Nokia's "Connecting People" has. In order to "understand documents", state of the art technologies are being used, which are discussed later in this post.

But first, let's understand the problem in depth: Why do we need to understand documents automatically? To gain control and efficiency over information, but also because understanding documents is the begining of building business knowledge.

In this way, information is like a full 500 Horsepower car. It's powerful, yet if you can't control this power it will surely control you and your life. Same applies to technology, wich by the way is always referred as being power  " Knowledge is Power " .

If you control information, you can increase efficiency managing CVs and reduce time from 8 minutes to 5 seconds, or reduce times in classification and manage of health documents by more than 90%.

And the best is that this new approach and enhancement over traditional document management software (which all compare with similar strengths) can be applied to many thriving scenarios. And with astonishing results in all of them.
So let's take a look now into the "How" this gets done, with a summary of State of the Art technologies that provide this type of breakthrough technology, inside areas of tremendous work such as machine learning and artificial intelligence (IA):

  • OCR: This technology has been widely used. Its its origins date back to 1914, but since that time, many improvements have been achieved in Optical Character Recognition technologies. Although its main goal remain converting images of printed text into machine readable text, developers have worked to improve the way we use OCR engines. At the beginning OCR helped the blind, today, OCR engines help another kind of blind: machines. Computers by themselves are unable to read images.  Assisted by OCR technologies, computers and ECM software are able to index content in PDFs documents (and other images) , and also, are able to find data within document contents. Lately, developers are working to find ways humans can interact with OCR engines, one of  their achievements is for example, visual template builders that enable users to design templates to extract data using zonal OCR. Athento Capture can provide this feature that makes incredibly easy to work with structured documents.

  • Semantic Technology: The goal of Semantics is to understand the meaning of words. If computers are able to understand words, they can build relations between them. This mean more capabilities when searching documents. Semantics makes it similar the way humans and machines do searches, and this is critical when having tons of information. Athento is using Semantics for its self-labeling feature called Autotagging. Thanks to Semantics, Athento is able to recognize relevant words within document content (terms which describe the document content) and turns them into labels that connect to other documents which include the same terms. This makes possible to find related documents very quick.

  • Machine Learning: This is a branch of Artificial Intelligence. Its main goal is that computers can act without tell them what to do explicitly. Developers are trying to get computers learning in the same way we do: practicing. Software which uses ML processes are able to improve their results by training. For example, Athento Capture compares documents using Neural Networks (Artificial Neural Networks -ANN), each time it receive a document type it improves its capabilities to recognize that kind of document. This is possible because  ANN technology finds patters within documents.  That explains why Athento Capture has the most high accuracy rate in document recognition in the market (98%). Another Machine Learning applications are Natural Language Processing (which Athento uses also), Search Engines, Speech recognition, etc.

  • Color Comparison: Besides other image analysis technologies, as ANN, Color Comparison is gaining ground as a technique of Image Processing. Particularly,  Color Histograms are making the difference in Color comparison methods. Its benefits are speed, and insensitivity to changes in camera viewpoint. We use CH to represent the distribution of colors in an image. In Document Recognition, if two images have a very similar color distribution, it is very probable that we are talking about the same kind of document. However, Color Histograms have its limitations. At Athento, we are using CH combined with other Image Processing techniques to improve the accuracy of document recognition feature.

LikeUs Yerbabuena Software on LinkedIn
Enhanced by Zemanta

No comments:

Post a Comment