Method of text information recognition from a graphical file with use of dictionaries and other supplementary data

• US 20080008386A1
• Filed: 07/06/2006
• Published: 01/10/2008
• Est. Priority Date: 07/06/2006
• Status: Active Grant

• Current Assignee: ABBYY Development LLC
• Original Assignee: ABBYY Software
• Inventors: Shamis, Alexander, Anisimovich, Konstantin, Rybkin, Vladimir
• Granted Patent: US 7,734,065 B2
• Time in Patent Office: Days
• US Class Current: 382/182
• CPC Class Codes: G06V 30/40 Document-oriented image-based pattern recognition

Abstract

The present invention deals with text comprising image parsed to graphemes. A result of character recognition is creation of one or more versions of characters for each grapheme. All possible words versions are obtained using all characters versions, and all parsing versions are examined. A supplementary data of several types is applied successively in the preliminarily prescribed order to the examined words. The processing with the use of supplemental data may be represented as a three times repeated processing of the same text fragment with the use of supplementary information becoming available at each time. The examination comprises three steps. 1) A set of chains LPG is built using all obtained recognized grapheme-to-character versions. 2) All obtained versions are analyzed with the successive application of subsequent supplemental data types in connection with the preliminarily assigned order or with a joint application thereof. 3) A supplementary space recognition correction.

Claims

1. A method of a text data recognition from an image file comprising

obtaining an image file from scanning device or from other source,

preliminarily assignment of the whole or a part of the following list of applied supplementary data types and an order of application thereto:

a line-to-graphemes parsing information and/or

a graphical element (grapheme) recognition quality, and/or

a whole words dictionary, and/or

a dictionary of permissible word fragments, and/or

rules, prescribed by applied standard data patterns or regular expressions, and/or

rules, prescribed by word disposition within the line or the paragraph, and/or

rules, prescribed by the document language peculiarities, and/or

rules, prescribed by the document type peculiarities, and/or

supplementary rules for rare occasions,

preliminarily assignment of an accuracy estimation for each type of supplementary data,

performance of one or more line-to-fragments parsing versions by reliably recognized spaces, said fragments presumably comprising single word images,

building of line partition graph (hereinafter LPG) for each line fragment, said graph describing fragment-to-graphemes parsing versions, said graphemes presumably comprising character images,

single graphemes recognition, using two or more classifiers of different types,

assignment of each said grapheme recognition version accuracy estimation,

interpretation of grapheme recognition version as a character version,

performance of at least the following steps;

the first step;

for each LPG chain connecting initial node and final node, a set of chains are built using all obtained recognized grapheme-to-character versions,

a total recognition accuracy level is calculated for each said chain,

obtained results are sorted in a total recognition accuracy descending order,

the second step;

all obtained character group versions are analyzed using supplemental information about capital-small characters disposition,

in a case of more than one grapheme-to-character recognition version being available, said each obtained recognition version is analyzed with the successive application of subsequent said supplemental data types in connection with the preliminarily assigned order or with a joint application thereof if necessary,

each obtained version is assigned an accuracy estimation,

character versions having said accuracy estimation lower, than the preliminarily assigned level are discarded,

the remain versions are sorted in a descending order using pair wise comparison;

the third step;

a supplementary space recognition correction is performed with respect to a previously mistakenly recognized spaces comprising;

joining of previously mistakenly separated elements,

separation of previously mistakenly combined elements.

2. The method as recited in claim 1, where said rules, prescribed by the document language peculiarities, comprises:

phonetic rules and/or

lexical rules, and/or

semantic rules.

3. The method as recited in claim 1, where said capital-small characters disposition comprises at least the following types:

all characters are capital characters,

all characters are small characters,

the first character is a capital character, and all the others are small,

version, based on the performed grapheme-to-character interpretations with the use of the said line-to-graphemes parsing information.

4. The method as recited in claim 1, where a particular dictionary of commonly used word fragments is applied.

5. The method as recited in claim 4, where each word fragment is provided with the application likelihood estimation thereof.

6. The method as recited in claim 4, where the dictionary of commonly used word fragments comprises all or a part of the following characters types restrictions:

bilingual word, and/or

bilingual word with numerals, and/or

dictionary identifier, and/or

abbreviation, and/or

number, and/or

Roman number, and/or

number with suffix, and/or

number with prefix, and/or

word composed of punctuators, and/or

word+number, and/or

number within a word, and/or

word including brackets, and/or

telephone number, and/or

URL pattern, and/or

filename with the full path, and/or

regular expressions pattern, and/or

supplementary pattern.

7. The method as recited in claim 1, further comprising rules for a new rules and/or restrictions addition, new data types rules addition including, said new data types further comprising simple and compound.

8. The method as recited in claim 7, where said compound data type further comprises a combination, of at least two components of simple type or any combination of simple and compound components.

9. The method as recited in claim 7, where said data type is assigned in a form of at least the following parameters:

a list of characters allowed for use in words and/or

a supplementary rule, restricting the allowed characters set and/or

a set of punctuators allowed to be used and/or

grammar rules for frequently used words or fragments thereof.