AutoCorrection
The most powerful feature of the TableExtractor is AutoCorrection. You can specify text templates for each column you have to extract. If the OCR process has made a mistake (e.g. by misinterpreting a 0 (zero) as an O (alpha numeric)) the tableextractor can correct that character.

Text patterns are described in QRX notation (similar to regular expressions). The following manual gives you an overview over QRX.

The QRX Manual
QRX offers an error tolerant way to match noisy or corrupted text structures against a specified text template. Therfore QRX uses a certain way of specifying text structures: The QRX notation. Like other forms of templates (e.g. regular expressions in Perl) it can be used to express your knowledge about a certain text information.

The following example may point out what is meant by this:

Assume we have an OCR generated text: My phone mumber is 030 2323. If we are searching for phone number with an conventional tool we won’t find anything because the is an m in the text instead of an n in the search text. With QRX we will find this match and we can decide if this error is acceptable.

Getting started: Matching basic strings

In the introduction we have matched two basic strings. This is quite simple to do with QRX:
 

QRX	Text to match	Error
phone number	phone mumber	1 character

The common character classes

In many cases we have knowledge about the type of characters (number, upper character, lower character, special character) which appear in the text. QRX offers a simple way to express that fact. There are 4 common classes:

Class	Usage
N	Numbers (0-9)
T	Upper letters
t	Lower letters
S	Special characters

Example: An US zip-code consists of 5 numerical characters ( “88123”).
 

The QRX syntax for one single number is “N”. Therefore a zip code can be expressed by “NNNNN”! 

QRX	Text to match	Error
NNNNN	88123	0 characters

Example: A date (german style) is written like “12.02.1976”. We use a combination of native elements (here “.”) with character classes:

QRX	Text to match	Error
NN.NN.NN	12.02.1976	0 characters

Another example:

QRX	Text to match	Error
My (first|second|third) beer	My second beEr	1 character

Alternatives

In QRX we have the ability to give several alternatives to a notation. The alternatives are separated by the character “|”. 

QRX	Text to match	Error
Jan|Feb|Mar|Apr	Fed	1 character

Sometimes it is necessary to assure that the OR-Separator relates to the right terms. Therefore brackets can be used. Example: We have dates where the year information can be consists of either 2 numbers or 4 numbers. Without brackets we get:

QRX	Text to match	Error
NN.NN.NN|NNNN	12.02.1976	2 characters

Using brackets we get:

QRX	Text to match	Error
NN.NN.(NN|NNNN)	12.02.1976	0 characters

Length Intervals

Up to this point all templates had fixed length. In many cases we have to match against information with a varying size but with known content (e.g. 5 to 10 upper letters or 3 to 5 numbers). With QRX we can use intervals after the character classes to express this fact. Therefore we define the minimum and the maximum size in curly brackets.

QRX	Text to match	Error
My t{5,12} beer	My ninth beer	0 characters

 
If there is more than one character class to find in this interval, we can use squared brackets to express that.

QRX	Text to match	Error
My number is [01]{1,10}	My number is 001x0001	1 character

The De-reference operator

As we already know there are existing several characters with a special meaning in the QRX syntax (“N”,”t”,”T”,”S”,”[“,”|”,..). If we want to use one of these characters as a native character, we have to insert the “@”-character directly before the relating letter.

Example: The upper “N” in number is interpreted as an number-class! The result is an 1 error distance! To solve this problem, use “@N”!

Annotation: This effect is not very common, in almost very case it is not necessary to look for those constellations!

QRX	Text to match	Error
myNumber	myNumber	1 characters! Because N is a symbol for numeric values!

With @-operator we get:

QRX	Text to match	Error
my@Number	myNumber	0 character

Notation

Expression	Match
x  @x     X|Y          (X)        X{n,m}    [abc]  N              T           t  S	The character x, if x is no spezial character. The character x (De-Reference) X or Y Group X with minimum n, maximum m repetitions (0 < n <= m). Characters a, b or c 0 to 9 (Numeric) A to Z (upper terminals) a to z (lower terminals) Special Characters: ASCII 1-31, 33-47, 58-64, 91-96, 123-255. Equals PERL notation: [^0-9A-Za-z\x7B-\x80]

Reserved Characters

QRX Version	Reserved Character
1.0	(){}[]|,NTtS@;

Comparison to regular expressions in PERL

QRX is a minor set of regular expressions. The following expressions are not available in QRX:

Not supported Perl Term	Description
X* X+  X? ^ $ . [a-z] \n \t \d	X any times X at least once X or none Start of Line End of Line Any Character Characters 'a' to 'z'. Note: In QRX notation:t. New line. Tab [0-9]. Note: In QRX notation:N.

 

Examples

QRX Expression	Examplematch
Hello?    @Telephone: N{3,5}/N{3,20}     Da@te: (N{1,2}.N{1,2}.N{2,4}|N{4,4}/N{2,2}/N{2,2})    Vorname: [Tt]{2,20}, @Nachname: [Tt]{2,20}.      Produc@t code: TTT-N{5,5}-[ABCDEFN]	Hello? Telephone: 0123/45678 Date: 2004/03/08 Vorname: Fritz, Nachname: Mueller. Product code: IBN-12345-3

QRX expressions in Backus-Naur

Term
QRX	Selection
Selection	Concartenation '|' Concartenation '|' Selection
Concartenation	'(' Selection ')' | Concartenation Concartenation | CharacterSpec
CharacterSpec	( '@' AnyCharacter | 'N' | 'T' | 't' | 'S' | CharacterClass | NonSpecialCharacter ) ( AreaSpec )?
CharacterClass	'[' ( CharacterClassItem )+ ']'
CharacterClassItem	'@' AnyCharacter | 'N' | 'T' | 't' | 'S' | NonSpecialCharacter
AreaSpec	'{' PositiveInteger ; PositiveInteger '}'
PositiveInteger	[0-9][0-9]*