Methods and systems for assessing the quality of automatically generated text

• US 8,682,648 B2
• Filed: 04/16/2013
• Issued: 03/25/2014
• Est. Priority Date: 02/05/2009
• Status: Active Grant

• Current Assignee: Google LLC (Alphabet Inc.)
• Original Assignee: Google Inc. (Alphabet Inc.)
• Inventors: Popat, Ashok
• Primary Examiner(s): AZAD, ABUL K
• Application Number:
  US13/864,180
• Publication Number:
  US 20130259378A1
• Time in Patent Office: 343 Days
• Field of Search: 704 1- 10, 382/291, 382/292
• US Class Current: 704/9
• CPC Class Codes: G06F 40/253 Grammatical analysis; Style critique
• CPC Class Codes: G06F 40/51 Translation evaluation
• CPC Class Codes: G06V 30/153 using recognition of characters or words
• CPC Class Codes: G06V 30/246 using linguistic properties, e.g. specific for English or German language
• CPC Class Codes: G06V 30/268 Lexical context

Abstract

A set of ordered characters is received in association with information specifying the locations of the characters within the image of the document. Language-conditional character probabilities for each character are determined based on a set of language models and the ordering of the characters. Neighbor characters associated with a target character are identified based on the locations of the characters. Language-conditional character probabilities associated with the neighbor characters and language-conditional character probabilities associated with the target character are combined to generate a local language-conditional likelihood associated with the target character, the local language-conditional likelihood representing a concordance of the target character to a language model.

Claims

1. A computer-implemented method of assessing quality of computer-generated text in a document image, the method comprising:

identifying text quality scores associated with a plurality of digital text characters generated from the document image, a text quality score for a target character describing a likelihood of the target character being at a location of the target character within the document image;

identifying a plurality of subsets of characters having associated text quality scores that differ from text quality scores associated with neighboring characters in the document image by more than a threshold value;

segmenting the document image into a plurality of segments associated with different text quality scores responsive to the identified plurality of subsets of characters;

determining a representative text quality score for each segment of the document image; and

storing the representative text quality scores in association with the segments.

2. The method of claim 1, wherein segmenting the document image in a plurality of segments comprises:

forming boundaries between a first subset of characters neighboring a second subset of characters in response to a difference between a first text quality score associated with the first subset of characters and a second text quality score associated with the second subset of characters exceeding the threshold value.

3. The method of claim 1, wherein identifying a subset of characters having associated text quality scores that differ from text quality scores associated with neighboring characters in the image comprises:

identifying neighboring characters in the document image having similar text quality scores; and

adding the neighboring characters in the document image having similar text quality scores to the subset of characters.

4. The method of claim 1, wherein identifying a subset of characters having associated text quality scores that differ from text quality scores associated with neighboring characters in the image comprises:

combining the text quality scores of the characters in the identified subset to produce a combined text quality score representing the identified subset of characters; and

wherein the segmenting segments the document image responsive to the combined text quality score.

5. The method of claim 1, wherein the threshold value defines a maximum difference of text quality scores for characters to be included in a given subset of characters.

6. The method of claim 5, further comprising adding a neighboring character to the given subset of characters in response to determining that a difference between an average value of text quality scores associated with the characters included in the given subset and a text quality score associated with the neighboring character subset is less than the threshold value.

7. The method of claim 1, wherein storing the representative text quality scores further comprises:

storing coordinates specifying a location in the document image of the segment associated with each representative text quality score.

8. A computer system for segmenting a document image, comprising:

a processor for executing computer program instructions;

a non-transitory computer-readable storage medium storing executable computer program instructions, the computer-readable storage medium comprising;

a score module configured to identify text quality scores associated with a plurality of digital text characters generated from the document image, a text quality score for a target character describing a likelihood of the target character being at a location of the target character within the document image;

a score analysis module configured to;

identify a plurality of subsets of characters having associated text quality scores that differ from text quality scores associated with neighboring characters in the document image by more than a threshold value;

segment the document image into a plurality of segments associated with different text quality scores responsive to the identified plurality of subsets of characters; and

determine a representative text quality score for each segment of the document image; and

a text quality database configured to store the representative text quality scores in association with the segments.

9. The system of claim 8, wherein segmenting the document image in a plurality of segments comprises:

forming boundaries between a first subset of characters neighboring a second subset of characters in response to a difference between a first text quality score associated with the first subset of characters and a second text quality score associated with the second subset of characters exceeding the threshold value.

10. The system of claim 8, wherein identifying a subset of characters having associated text quality scores that differ from text quality scores associated with neighboring characters in the image comprises:

identifying neighboring characters in the document image having similar text quality scores; and

adding the neighboring characters in the document image having similar text quality scores to the subset of characters.

11. The system of claim 8, wherein identifying a subset of characters having associated text quality scores that differ from text quality scores associated with neighboring characters in the image comprises:

combining the text quality scores of the characters in the identified subset to produce a combined text quality score representing the identified subset of characters; and

wherein the segmenting segments the document image responsive to the combined text quality score.

12. The system of claim 8, wherein the threshold value defines a maximum difference of text quality scores for characters to be included in a given subset of characters.

13. The system of claim 12, further comprising adding a neighboring character to the given subset of characters in response to determining that a difference between an average value of text quality scores associated with the characters included in the given subset and a text quality score associated with the neighboring character subset is less than the threshold value.

14. The system of claim 8, wherein storing the representative text quality scores further comprises:

storing coordinates specifying a location in the document image of the segment associated with each representative text quality score.

15. A non-transitory computer-readable storage medium storing executable computer program instructions for performing steps comprising:

identifying text quality scores associated with a plurality of digital text characters generated from the document image, a text quality score for a target character describing a likelihood of the target character being at a location of the target character within the document image;

identifying a plurality of subsets of characters having associated text quality scores that differ from text quality scores associated with neighboring characters in the document image by more than a threshold value;

segmenting the document image into a plurality of segments associated with different text quality scores responsive to the identified plurality of subsets of characters;

determining a representative text quality score for each segment of the document image; and

storing the representative text quality scores in association with the segments.

16. The storage medium of claim 15, wherein segmenting the document image in a plurality of segments comprises:

forming boundaries between a first subset of characters neighboring a second subset of characters in response to a difference between a first text quality score associated with the first subset of characters and a second text quality score associated with the second subset of characters exceeding the threshold value.

17. The storage medium of claim 15, wherein identifying a subset of characters having associated text quality scores that differ from text quality scores associated with neighboring characters in the image comprises:

identifying neighboring characters in the document image having similar text quality scores; and

adding the neighboring characters in the document image having similar text quality scores to the subset of characters.

18. The storage medium of claim 15, wherein identifying a subset of characters having associated text quality scores that differ from text quality scores associated with neighboring characters in the image comprises:

combining the text quality scores of the characters in the identified subset to produce a combined text quality score representing the identified subset of characters; and

wherein the segmenting segments the document image responsive to the combined text quality score.

19. The storage medium of claim 15, wherein the threshold value defines a maximum difference of text quality scores for characters to be included in a given subset of characters.

20. The storage medium of claim 19, further comprising adding a neighboring character to the given subset of characters in response to determining that a difference between an average value of text quality scores associated with the characters included in the given subset and a text quality score associated with the neighboring character subset is less than the threshold value.

21. The storage medium of claim 15, wherein storing the representative text quality scores further comprises:

storing coordinates specifying a location in the document image of the segment associated with each representative text quality score.