Segmental rescoring in text recognition
First Claim
1. A method for text recognition of a pixelated image with unknown text in a first region of said image, the method comprising:
- generating a plurality text hypotheses, each text hypothesis representing the unknown text in the first region, each text hypothesis being associated with a corresponding score;
for each text hypothesis of the generated hypotheses, forming data representing one or more segmentations of the first region of the image according to the hypothesis, each segmentation including a series of segments of the image, each segment corresponding to a part of the text hypothesis;
for each of the one or more segmentations, for each segment in the segmentation, forming separate data representing segmental features of the segment;
determining a segmental score for each segment according to the segmental features of the segment and the corresponding part of the text hypothesis associated with the segmentation including the segment;
for each text hypothesis, determining an overall segmental score according to the determined segmental score for the segments of the one or more segmentations associated with the text hypothesis, and determining an overall score by combining the overall segmental score and the corresponding score associated with the hypotheses; and
providing data representing a text recognition the first region of the image according to the determined overall score for each of the generated text hypotheses for the image.
2 Assignments
0 Petitions
Accused Products
Abstract
A method for text recognition includes generating a number of text hypotheses for an image, for example, using an HMM based approach using fixed-width analysis features. For each text hypothesis, one or more segmentations are generated and scored at the segmental level, for example, according to character or character group segments of the text hypothesis. In some embodiments, multiple alternative segmentations are considered for each text hypothesis. In some examples, scores determined in generating the text hypothesis and the segmental score are combined to select an overall text recognition of the image.
-
Citations
34 Claims
-
1. A method for text recognition of a pixelated image with unknown text in a first region of said image, the method comprising:
-
generating a plurality text hypotheses, each text hypothesis representing the unknown text in the first region, each text hypothesis being associated with a corresponding score; for each text hypothesis of the generated hypotheses, forming data representing one or more segmentations of the first region of the image according to the hypothesis, each segmentation including a series of segments of the image, each segment corresponding to a part of the text hypothesis; for each of the one or more segmentations, for each segment in the segmentation, forming separate data representing segmental features of the segment; determining a segmental score for each segment according to the segmental features of the segment and the corresponding part of the text hypothesis associated with the segmentation including the segment; for each text hypothesis, determining an overall segmental score according to the determined segmental score for the segments of the one or more segmentations associated with the text hypothesis, and determining an overall score by combining the overall segmental score and the corresponding score associated with the hypotheses; and providing data representing a text recognition the first region of the image according to the determined overall score for each of the generated text hypotheses for the image. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20)
-
-
21. A text recognition system for text recognition of a pixelated image with unknown text in a first region of said image, the system comprising:
-
a first text recognition system configured to generate a plurality text hypotheses, each text hypothesis representing the unknown text in the first region, each text hypothesis being associated with a corresponding score, the first recognition system being further configured, for each text hypothesis of the generated hypotheses, to form data representing one or more segmentations of the first region of the image according to the hypothesis, each segmentation including a series of segments of the image, each segment corresponding to a part of the text hypothesis; a segment processor configured to accept the generated text hypotheses and associated segmentations from the first recognition system, and, for each text hypothesis, form one or more segmentations of the image associated with the hypothesis, each segmentation including a series of segments of the image, each segment corresponding to a part of the text hypothesis, and for each of the one or more segmentations, for each segment in the segmentation, forming separate data representing segmental features of the segment; wherein the segment processor includes a segment scorer for determining a segmental score for each segment according to the segmental features of the segment and the corresponding part of the text hypothesis associated with the segmentation including the segment; wherein the segment processor is further configured, for each text hypothesis, to determine an overall segmental score according to the determined segmental score for the segments of the one or more segmentations associated with the text hypothesis; the system further comprising a scorer configured, for each text hypothesis, to determine an overall score by combining the overall segmental score and the corresponding score generated by the first text recognition system, and to output data representing a text recognition the first region of the image according to the determined overall score for each of the generated text hypotheses for the image. - View Dependent Claims (22, 23)
-
-
24. Software instructions embodied on a non-transitory computer readable medium for causing a data processing system to:
-
generate a plurality text hypotheses, each text hypothesis representing unknown text in a first region of a pixelated image that includes text, each text hypothesis being associated with a first score; for each text hypothesis of the generated hypotheses, form data representing one or more segmentations of the first region of the image according to the hypothesis, each segmentation including a series of segments of the image, each segment corresponding to a part of the text hypothesis; for each of the one or more segmentations, for each segment in the segmentation, form separate data representing segmental features of the segment; determine a segmental score for each segment according to the segmental features of the segment and the corresponding part of the text hypothesis associated with the segmentation including the segment; for each text hypothesis, determine an overall segmental score according to the determined segmental score for the segments of the one or more segmentations associated with the text hypothesis, and determine an overall score by combining the overall segmental score and the first score associated with the hypotheses; and provide data representing a text recognition the first region of the image according to the determined overall score for each of the generated text hypotheses for the image. - View Dependent Claims (25, 26)
-
-
27. A computer-implemented method for text recognition of an optically acquired image, the method comprising:
-
accepting data representing a region of an image containing an unknown text; using a first text recognition procedure to process the accepted data, including identifying a set of character sequences that hypothetically represent the unknown text and identifying variable width segments in the image, each variable width segment corresponding to a character in a character sequence of the set of character sequences; computing, for each segment of the identified variable width segments, one or more segmental features from the portion of the image associated with that segment; and using the computed segmental features to determine, for at least some character sequences of the set of character sequences identified using the first text recognition procedure, a recognition score for said character sequence. - View Dependent Claims (28, 29, 30, 31, 32, 33, 34)
-
Specification