Back to results
US 8,280,157 B2Granted
Methods and systems for refining text detection in a digital image
AssigneeSharp Laboratories of America, Inc.
InventorsLawrence Shao-hsien Chen, Jon M. Speigle, Ahmet Mufit Ferman, Richard John Campbell
Life of this patent
Estimating the term…
Abstract
Embodiments of the present invention comprise systems and methods for refining text-detection results for a digital image.
Claims22
A method for classifying a pixel in a digital image, said method comprising:
- receiving a first classification map, wherein said first classification map identifies candidate pixels belonging to a first class in a digital image;
- receiving a first pixel location in said first classification map, wherein said first pixel location in said first classification map corresponds to a first pixel location in said digital image;
- calculating a first color estimate corresponding to said first pixel location in said digital image, wherein said first color estimate is a summary of the colors of a plurality of pixels belonging to said first class located in a region proximate to said first pixel location in said digital image;
- determining a validity state associated with said first color estimate, wherein said validity state is determined to be a state associated with an invalid first color estimate when a second plurality of pixels associated with said region does not contain at least one pixel identified as belonging to said first class according to said first classification map;
- calculating a first color difference from said first color estimate and the color of the pixel in said digital image at said first pixel location in said digital image;
- and
- refining said first classification map at said first pixel location in said first classification map based on said first color difference and said validity state.
4 more elements
A system for classifying a pixel in a digital image, said system comprising:
- a classification map receiver, in a computing device, for receiving a first classification map, wherein said first classification map identifies candidate pixels belonging to a first class in a digital image;
- a location receiver, in said computing device, for receiving a first pixel location in said first classification map, wherein said first pixel location in said first classification map corresponds to a first pixel location in said digital image;
- a first color estimate calculator, in said computing device, for calculating a first color estimate corresponding to said first pixel location in said digital image, wherein said first color estimate is a summary of the colors of a plurality of pixels belonging to said first class located in a region proximate to said first pixel location in said digital image;
- a validity determiner, in said computing device, for determining a validity state associated with said first color estimate, wherein said validity state is determined to be a state associated with an invalid first color estimate when a second plurality of pixels associated with said region does not contain at least one pixel identified as belonging to said first class according to said first classification map;
- a first color difference calculator, in said computing device, for calculating a first color difference from said first color estimate and the color of the pixel in said digital image at said first pixel location in said digital image;
- and
- a first map refiner, in said computing device, for refining said first classification map at said first pixel location in said first classification map based on said first color difference and said validity state.
4 more elements
A method for classifying a pixel in a digital image, said method comprising:
- receiving a text map, wherein said text map identifies candidate text pixels in a digital image;
- receiving a first pixel location in said text map, wherein said first pixel location in said text map corresponds to a first pixel location in said digital image;
- calculating a text-color estimate corresponding to said first pixel location in said digital image, wherein said text-color estimate is a summary of the colors of a plurality of pixels belonging to a text class, as indicated by said text map, located in a region proximate to said first pixel location in said digital image;
- determining a validity state associated with said text-color estimate, wherein said validity state is determined to be a state associated with an invalid text-color estimate when a second plurality of pixels associated with said region does not contain at least one pixel identified as belonging to a text class, as indicated by said text map;
- calculating a text-color difference from said text-color estimate and the color of the pixel in said digital image at said first pixel location in said digital image;
- calculating a background-color estimate corresponding to said first pixel location in said digital image, wherein said background-color estimate is a summary of the colors of a plurality of pixels not belonging to said text class, as indicated by said text map, location in a region proximate to said first pixel location in said digital image;
- calculating a background-color difference from said background-color estimate and the color of the pixel in said digital image at said first pixel location in said digital image;
- refining said pixel at said first pixel location in said text map as a text pixel if said text-color difference is less than a first threshold and said pixel at said first pixel location in said text map is identified as non-text and said validity state associated with said text-color estimate is not said state associated with an invalid text-color estimate;
- refining said pixel at said first pixel location in said text map a non-text pixel if said background-color difference is less than a second threshold and said pixel at said first pixel location in said text map is identified as text;
- and
- wherein said summary of the colors of said plurality of pixels, belonging to said text class, located in said region proximate to said first pixel location in said digital image comprises a calculation selected from the group consisting of a trimmed-mean of said colors of said plurality of pixels, belonging to said text class, located in said region proximate to said first pixel location in said digital image, a median of said colors of said plurality of pixels, belonging to said text class, located in said region proximate to said first pixel location in said digital image and a weighted average of said colors of said plurality of pixels, belonging to said text class, located in said region proximate to said first pixel location in said digital image.
8 more elements
Dependencies are inferred from the claim text.