Вы находитесь на странице: 1из 2

An Overview on Automated Form Recognition

What is Handprint (ICR) Recognition ? Handprint recognition, sometimes called I.C.R., for Intelligent Character Recognition, is a process where Hand Printed Alpha Numeric Characters are interpreted through a software engine that compares the bitmapped image of the character to a large sampling (1000s) of actual hand printed characters and makes an intelligent decision as to what the character represents. The software then interprets each remaining character in the field to make a complete string (i.e. a Part Number or Name). The engines results are controlled by many factors including: Confidence Thresholds - Most Handprint Engines have a default setting (typically 80%) that is used a minimum level that the engines must exceed during the interpretation process. If the Engines Confidence is below the setting on one or more characters, the character(s) are highlighted during the Review Process and a Human Operator reviews the character(s) and makes any corrections. This setting can also be adjusted by the user on a global (form level) or field by field basis to match the requirements of a specific application. Database Lookup/Dictionary Matching - Through matching the results of an interpreted field against an existing database or data dictionary, invalid data cannot be written to the output file. If the field does not match a value in the database Character for Character, the field is highlighted during the Review Process. Entry Required - If the interpreted field contains no data, the field is highlighted during the Review Process. This ensures that required Key Index Data Fields must be filled in. For example, a Social Security Number or Account Number. Range Check - In a numeric field, the interpreted results must fall inclusively between the range. For example, the Month in a Date field must be between 1 and 12 or a Quantity Field must be between 1 and 25 Units. Field Template (Mask) - Each position in a field can be set up to interpret a specific type of character. For example, a Part Number is always two (2) Alpha Characters followed by Five (5) Numeric Characters. The Fields template would be set up as AANNNNN. Conditional Branching - If a Field is responded to in a certain way, one or more other fields can be checked for subsequent responses. For example, If Question 1 is answered Yes, Question 2 must also be filled in. Missing Pages - If a multiple page form utilizes a field with a sequentially numbered bar code or pre-printed number on each page, the software will not allow an incomplete data record to be written and will ensure that a set missing a page cannot be combined with a another sets page. What Accuracy can I expect from Handprint Recognition ? Handprint Recognition has improved dramatically over the last two (2) years. Todays Voting Engines (see below) are up to 250% more accurate than single engines. Numeric Recognition is the most accurate since there is only Ten (10) characters to differentiate between. Numeric Recognition can achieve 95% first pass read rates, Alpha 90% and Alpha Numeric 85%. What is a Voting Engine ?

A Voting Engine is Two (2) or more Handprint or OCR recognition engines. Each engine provides its highest confidence character to the Voting Layer. The Voting Layer Algorithm takes the responses and weights each one according to the confidence level returned and the known strengths and attributes of the engine. A simple analogy is two heads are better than one. What is Machine Print (OCR) Recognition ? O.C.R., for Optical Character Recognition, is a process where Machine Printed (Laser, Dot Matrix) Alpha Numeric Characters are interpreted through a software engine that compares the bitmapped image of the character to a large sampling of actual machine printed characters and makes an intelligent decision as to what the character represents. The software then interprets each remaining character in the field on a form to make a complete string (i.e. a Part Number). The engines accuracy is controlled by a variable confidence threshold, through matching the results against an existing database or data dictionary and all of the other Constraints discussed in Handprint Recognition. What is Shaded Circle (Optical Mark) Recognition ? Optical Mark Recognition, referred to as OMR, is a process where the software determines whether a response has been entered based on what amount of the interior area of a circle or box was filled in by the user. This very accurate method of data capture is used in surveys, testing and many other applications. Todays Software no longer requires the use of Number 2 Lead Pencils! What other Recognition can these types of Software offer? Bar Code - Most packages interpret horizontally or vertically printed or affixed Bar Codes (Code 39, Code 128, I 2 of 5, UPC, EAN and Codabar. Some packages even interpret two (2) dimensional PDF-417 bar codes. These extremely dense codes permit up to 2,000 bytes of data in a single code block. Merged Data Fields - By using an external database file, many key fields can have data merged and printed onto the form and interpreted along with the User entered fields. This process, virtually identical to your Word Processors capability, provides an extremely high first pass read rate and reduces the amount of manual data verification and correction. How long does it take to interpret a Form ? A single page form can take anywhere from 4 to 10 seconds to interpret (based on the complexity and type of data interpretation, workstation processor speed, memory, etc.) How much Manual Data Entry time can I expect to save ? A Conservative Sixty-Five Percent (65%) or more.

Вам также может понравиться