Method for processing optical character recognizer output

• US 8,983,211 B2
• Filed: 05/14/2012
• Issued: 03/17/2015
• Est. Priority Date: 05/14/2012
• Status: Active Grant

• Current Assignee: Xerox Corporation (Xerox Holdings Corp.)
• Original Assignee: Xerox Corporation (Xerox Holdings Corp.)
• Inventors: Venkatapathy, Sriram, Cancedda, Nicola
• Primary Examiner(s): Carter, Aaron W
• Application Number:
  US13/470,434
• Publication Number:
  US 20130301920A1
• Time in Patent Office: 1,037 Days
• Field of Search: 382182-185, 382229-231, 382/321
• US Class Current: 382/229
• CPC Class Codes: G06F 40/232 Orthographic correction, e.g. spell checking or vowelisation
• CPC Class Codes: G06F 40/58 Use of machine translation, e.g. for multi-lingual retrieval, for server-side translation for client devices or for real-time translation
• CPC Class Codes: G06V 30/10 Character recognition
• CPC Class Codes: G06V 30/12 Detection or correction of errors, e.g. by rescanning the pattern
• CPC Class Codes: G06V 30/262 using context analysis, e.g. lexical, syntactic or semantic context

Abstract

A method, a system, and a computer program product for processing the output of an OCR are disclosed. The system receives a first character sequence from the OCR. A first set of characters from the first character sequence are converted to a corresponding second set of characters to generate a second character sequence based on a look-up table and language scores.

Claims

1. A computer implemented method for processing an output of an optical character recognizer (OCR), the computer implemented method comprising:

receiving an OCR-converted character sequence from the OCR; and

converting a set of OCR-converted characters of the OCR-converted character sequence to a corresponding corrected set of characters to generate a corrected character sequence based on;

a probability of replacing an OCR-converted character of the set of OCR-converted characters with one or more target characters to form one or more characters of the corrected set of characters, and

language scores generated by a language model.

2. The computer implemented method of claim 1, further comprising scanning a document to generate an electronic document, wherein the electronic document is optically recognized by the OCR to generate the OCR-converted character sequence.

3. The computer implemented method of claim 1, further comprising optically recognizing characters from an electronic document to generate the OCR-converted character sequence.

4. The computer implemented method of claim 1, wherein a list of OCR-converted characters and the corresponding one or more target characters is stored in a look-up table.

5. The computer implemented method of claim 1, wherein the converting comprises implanting a finite state transducer (FST) corresponding to each character of the OCR-converted character sequence.

6. The computer implemented method of claim 5, further comprising assigning weights to the FST and the language model using a Minimum Error Rate Training (MERT) technique.

18. The computer implemented method of claim 5, wherein the FST corresponding to each character of the OCR-converted character sequence is associated with a plurality of costs corresponding to a plurality of possible replacements of the OCR-converted character of the set of OCR-converted characters to the one or more target characters.

19. The computer implemented method of claim 18, wherein each of the plurality of costs represents a value such that a likelihood of replacing the OCR converted character of the set of OCR-converted characters with the one or more target characters is inversely proportional to the value of the corresponding cost.

7. A computer implemented method for processing an output of an optical character recognizer (OCR), the computer implemented method comprising:

receiving a first character sequence from the OCR,

converting a first set of characters from the first character sequence to a corresponding second set of characters to generate a second character sequence based on one or more finite state transducers (FSTs) corresponding to each character of the first character sequence and language scores generated by a language model, wherein weights associated with the one or more FSTs and the language model are determined using a Minimum Error Rate Training (MERT) technique, wherein a number of edits for the conversion of the first set of characters to the corresponding second set of characters is less than or equal to a predetermined number of edits.

8. A computer implemented method for language translation comprising:

receiving an Optical Character Recognizer (OCR)-converted character sequence from an OCR, the OCR-converted character sequence being in a first language;

converting a set of OCR-converted characters of the OCR-converted character sequence to a corresponding corrected set of characters to generate a corrected character sequence based on a look-up table and language scores generated by a language model, wherein the corrected character sequence is in the first language, wherein the look-up table comprises a probability score for each character of the set of OCR-converted characters, wherein the probability score represents a probability of replacing an OCR-converted character of the set of OCR-converted characters with one or more target characters to form one or more characters of the corrected set of characters; and

translating an OCR-converted word sequence corresponding to the corrected character sequence to a corrected word sequence in a second language.

15. The computer implemented method of claim 8, wherein the look-up table is generated by extracting a plurality of mappings of each character of the set of OCR-converted characters with the one or more target characters to compute and assign the probability score to each of the extracted plurality of mappings.

16. The computer implemented method of claim 8, wherein the language scores are based on a probability of a character in the corrected character sequence following a pre-determined sequence of one or more characters.

17. The computer implemented method of claim 16, wherein the language scores are based on a product of the probability of each character in the corrected character sequence following the pre-determined sequence of one or more characters for the entire corrected character sequence.

9. A system for processing an output of an optical character recognizer (OCR), the system comprising a processor coupled to a memory, the memory having stored therein one or more program modules comprising:

a conversion module configured to convert an OCR-converted set of characters in an OCR-converted character sequence to a corresponding corrected set of characters to generate a corrected character sequence based on a look-up table and language scores, wherein the OCR-converted character sequence is received from the OCR, wherein the look-up table comprises a probability score for each character of the set of OCR-converted characters, wherein the probability score represents a probability of replacing an OCR-converted character of the set of OCR-converted characters with one or more target characters to form one or more characters of the corrected set of characters.

10. The system of claim 9, wherein the OCR is configured to optically recognize characters from an electronic document to generate the OCR-converted character sequence.

11. The system of claim 9, wherein the look-up table comprises a list of OCR-converted characters and the corresponding one or more target characters.

12. The system of claim 9 wherein, the conversion module is configured to convert the OCR-converted set of characters to the corrected set of characters based on a cost, wherein the cost corresponds to a value indicating a likelihood of the OCR-converted set of characters of the OCR-converted character sequence being the corrected set of characters.

13. The system of claim 9, further comprising a language model configured to calculate the language scores.

20. The system of claim 9, further comprising a training module configured to generate the look-up table by extracting a plurality of mappings of each character of the set of OCR-converted characters with the one or more target characters to compute and assign the probability score to each of the extracted plurality of mappings.

14. A system for language translation, the system comprising a processor coupled to a memory, the memory having stored therein one or more program modules comprising:

a conversion module configured to convert an Optical Character Recognizer (OCR)-converted set of characters in an OCR-converted character sequence to a corresponding corrected set of characters to generate a corrected character sequence based on a look-up table and language scores generated by a language model, the OCR-converted character sequence being received from an OCR, wherein the OCR-converted character sequence and the character sequence are in a first language, wherein the look-up table comprises a probability score that represents a probability of replacing an OCR-converted character of the set of OCR-converted characters with one or more target characters to form one or more characters of the corrected set of characters; and

a translation module configured to translate an OCR-converted word sequence corresponding to the corrected character sequence to a corrected word sequence in a second language.