Methods and systems for identifying text orientation in a digital image

• US 8,144,989 B2
• Filed: 06/21/2007
• Issued: 03/27/2012
• Est. Priority Date: 06/21/2007
• Status: Active Grant

• Current Assignee: Sharp Kabushiki Kaisha (Hon Hai Precision Industry Co., Ltd.)
• Original Assignee: Sharp Laboratories of America Incorporated (Hon Hai Precision Industry Co., Ltd.)
• Inventors: Speigle, Jon M., Ferman, Ahmet Mufit
• Primary Examiner(s): Carter, Aaron W
• Application Number:
  US11/766,661
• Publication Number:
  US 20080317341A1
• Time in Patent Office: 1,741 Days
• Field of Search: 382/168, 382177-188, 382/291, 715/204
• US Class Current: 382/177
• CPC Class Codes: G06V 10/242 by image rotation, e.g. by 90 degrees

Abstract

Aspects of the present invention relate to systems and methods for determining text orientation in a digital image.

Claims

1. A method for determining a text orientation in a digital image, said method comprising:

computing, for each of a plurality of text-object pairs in a digital image, a first alignment feature value for a first alignment feature, thereby producing a first plurality of feature values;

computing, for each of said plurality of text-object pairs, a second alignment feature value for a second alignment feature, thereby producing a second plurality of feature values;

generating a first histogram of said first plurality of feature values;

generating a second histogram of said second plurality of feature values;

determining a text orientation in a digital image based on said first histogram and said second histogram; and

wherein each of said plurality of text-object pairs comprises a first text object and a second text object; and

;

said computing, for each of said plurality of text-object pairs, said first alignment feature value for said first alignment feature comprises;

determining a first absolute difference between a first bounding edge of said first text object and an associated first bounding edge of said second text object, thereby associating said first histogram with said first bounding edge; and

said computing, for each of said plurality of text-object pairs, said second alignment feature value for said second alignment feature comprises;

determining a second absolute difference between a second bounding edge of said first text object and an associated second bounding edge of said second text object, thereby associating said second histogram with said second bounding edge.

2. A method according to claim 1, wherein said determining said text orientation comprises:

determining a first frequency-of-occurrence from said first histogram, wherein said first frequency-of-occurrence corresponds to a substantially zero first absolute difference;

determining a second frequency-of-occurrence from said second histogram, wherein said second frequency-of-occurrence corresponds to a substantially zero second absolute difference;

associating a baseline-side of said text orientation with said first bounding edge when said first frequency-of-occurrence is greater than said second frequency-of-occurrence and text characters with ascenders occur more frequently than text characters with descenders in a written language; and

associating said baseline-side of said text orientation with said second bounding edge when said first frequency-of-occurrence is less than said second frequency-of-occurrence and text characters with ascenders occur more frequently than text characters with descenders in said written language.

3. A method according to claim 1, wherein said determining said text orientation comprises:

determining a first frequency-of-occurrence from said first histogram, wherein said first frequency-of-occurrence corresponds to a substantially zero first absolute difference;

determining a second frequency-of-occurrence from said second histogram, wherein said second frequency-of-occurrence corresponds to a substantially zero second absolute difference;

associating said baseline-side of said text orientation with said first bounding edge when said first frequency-of-occurrence is less than said second frequency-of-occurrence and text characters with descenders occur more frequently than text characters with ascenders in a written language; and

associating said baseline-side of said text orientation with said second bounding edge when said first frequency-of-occurrence is greater than said second frequency-of-occurrence and text characters with descenders occur more frequently than text characters with ascenders in said written language.

4. A method for determining a text orientation in a digital image, said method comprising:

computing, for each of a plurality of text-object pairs in a digital image, a first alignment feature value for a first alignment feature, thereby producing a first plurality of feature values;

computing, for each of said plurality of text-object pairs, a second alignment feature value for a second alignment feature, thereby producing a second plurality of feature values;

generating a first histogram of said first plurality of feature values;

generating a second histogram of said second plurality of feature values;

determining a text orientation in a digital image based on said first histogram and said second histogram;

computing, for each of said plurality of text-object pairs, a third alignment feature value for a third alignment feature, thereby producing a third plurality of feature values;

computing, for each of said plurality of text-object pairs, a fourth alignment feature value for a fourth alignment feature, thereby producing a fourth plurality of feature values

generating a third histogram of said third plurality of feature values;

generating a fourth histogram of said fourth plurality of feature values; and

wherein said determining said text orientation in said digital image based on said first histogram, said second histogram, said third histogram, said fourth histogram.

5. A method according to claim 4, wherein each of said plurality of text-object pairs comprises a first text object and a second text object;

and;

said computing, for each of said plurality of text-object pairs, said third alignment feature value for said third alignment feature comprises;

determining a third absolute difference between a third bounding edge of said first text object and an associated third bounding edge of said second text object, thereby associating said third histogram with said third bounding edge; and

said computing, for each of said plurality of text-character pairs, said fourth alignment feature value for said fourth alignment feature comprises;

determining a fourth absolute difference between a fourth bounding edge of said first text object and an associated fourth bounding edge of said second text object, thereby associating said fourth histogram with said fourth bounding edge.

6. A method according to claim 5, wherein said determining said text orientation comprises:

determining a first frequency-of-occurrence from said first histogram, wherein said first frequency-of-occurrence corresponds to a substantially zero first absolute difference;

determining a second frequency-of-occurrence from said second histogram, wherein said second frequency-of-occurrence corresponds to a substantially zero second absolute difference;

determining a third frequency-of-occurrence from said third histogram, wherein said third frequency-of-occurrence corresponds to a substantially zero third absolute difference;

determining a fourth frequency-of-occurrence from said fourth histogram, wherein said fourth frequency-of-occurrence corresponds to a substantially zero fourth absolute difference;

determining a maximum frequency-of-occurrence of said first frequency-of-occurrence, said second frequency-of-occurrence, said third frequency-of-occurrence and said fourth frequency-of-occurrence, thereby determining a histogram associated with said maximum frequency-of-occurrence; and

associating a baseline-side of said text orientation with said bounding edge associated with said histogram associated with said maximum frequency-of-occurrence when text characters with ascenders occur more frequently than text characters with descenders in a written language.

7. A method according to claim 5, wherein said determining said text orientation comprises:

determining a first frequency-of-occurrence from said first histogram, wherein said first frequency-of-occurrence corresponds to a substantially zero first absolute difference;

determining a second frequency-of-occurrence from said second histogram, wherein said second frequency-of-occurrence corresponds to a substantially zero second absolute difference;

determining a third frequency-of-occurrence from said third histogram, wherein said third frequency-of-occurrence corresponds to a substantially zero third absolute difference;

determining a fourth frequency-of-occurrence from said fourth histogram, wherein said fourth frequency-of-occurrence corresponds to a substantially zero fourth absolute difference;

determining a maximum frequency-of-occurrence of said first frequency-of-occurrence, said second frequency-of-occurrence, said third frequency-of-occurrence and said fourth frequency-of-occurrence, thereby determining a histogram associated with said maximum frequency-of-occurrence; and

associating a baseline-side of said text orientation with a bounding edge opposite of said bounding edge associated with said histogram associated with said maximum frequency-of-occurrence when text characters with descenders occur more frequently than text characters with ascenders in a written language.

8. A method for determining a text orientation in a digital image, said method comprising:

selecting a first text object from a digital image;

determining, for said first text object, a first bounding box;

selecting a second text object from said digital image;

determining, for said second text object, a second bounding box;

when the size of said first text object is substantially different than the size of said second text object,

determining, for said first text object and said second text object, a first bounding-box feature value for a first bounding-box feature, wherein said first bounding-box feature is related to a first edge of said first bounding box;

updating a first histogram based on said first bounding-box feature value;

determining, for said first text object and said second text object, a second bounding-box feature value for a second bounding-box feature, wherein said second bounding-box feature is related to a second edge of said first bounding box;

updating a second histogram based on said second bounding-box feature value;

determining, for said first text object and said second text object, a third bounding-box feature value for a third bounding-box feature, wherein said third bounding-box feature is related to a third edge of said first bounding box;

updating a third histogram based on said third bounding-box feature value;

determining, for said first text object and said second text object, a fourth bounding-box feature value for a fourth bounding-box feature, wherein said fourth bounding-box feature is related to a fourth edge of said first bounding box; and

updating a fourth histogram based on said fourth bounding-box feature value; and

determining a text orientation in said digital image based on said first histogram, said second histogram, said third histogram and said fourth histogram.

9. A method according to claim 8, wherein said first text object and said second text object are adjacent text objects in said digital image.

10. A method according to claim 8, wherein said first text object and said second text object are not adjacent text objects in said digital image.