Method and apparatus for image processing employing image segmentation using tokenization
AssigneeXerox Corporation
InventorsWilliam J. Rucklidge, Daniel P. Huttenlocher, Pedro Felzenswalb
Life of this patent
Estimating the term…
Abstract
A method for image processing including segmenting document images into text and continuous tone regions is provided. The method uses a token shape comparison and resultant data (e.g. shape recurrence) to determine which high-contrast regions of the image correspond to text or other symbols. The identified text is then separated from the image and processing ensues.
Claims14
A method for processing an image including text and continuous tone regions, the method comprising the steps of:
- performing a thresholding routine on the image to transform the image to a second image comprising components representing high contrast regions of the image;
- determining which components have connected black pixels to obtain a first group of components;
- determining which components have connected white pixels to obtain a second group of components;
- inverting one of the first and second groups of components to obtain a third group of components including both groups;
- determining a color of each component based on the image;
- determining a variance of color of each component;
- selecting a fourth group of components from the third group based on the determination of the variance of color whereby components having variance of color under a predetermined threshold are selected;
- calculating Hamming distances between scan lines of each component of the fourth group;
- calculating an average Hamming distance for each component of the fourth group;
- selecting a fifth group of components based on the average Hamming distances whereby components having an average Hamming distance under a predetermined threshold are selected;
- determining which components align with other components;
- selecting a sixth group of components from the fifth group based on the determination of alignment;
- classifying the components of the sixth group according to shape to obtain classes of components;
- identifying components of the sixth group belonging to a class of more than one component as text;
- identifying unidentified components of the sixth group that align with components of similar color as text;
- identifying unidentified components of the sixth group that align with the identified text as text;
- and,
- identifying unidentified components of the sixth group that are in close proximity to the identified text and relatively small in comparison to the identified text as text.
A method for processing an image including text and continuous tone regions, the method comprising the steps of:
- performing a thresholding routine on the image to transform the image to a second image comprising components representing high contrast regions of the image;
- selecting a group of the components based on color variances, Hamming distances and alignment;
- classifying the components of the selected group according to shape to obtain classes of components;
- identifying components of the selected group belonging to a class of more than one component as text;
- determining which components have connected black pixels to obtain a first group of components;
- determining which components have connected white pixels to obtain a second group of components;
- inverting one of the first and second groups of components to obtain a third group of components including both groups;
- determining a color of each component based on the image;
- determining a variance of color of each component;
- selecting a fourth group of components from the third group based on the determination of the variance of color whereby components having variance of color under a predetermined threshold are selected;
- calculating Hamming distances between scan lines of each component of the fourth group;
- calculating an average Hamming distance for each component of the fourth group;
- and,
- selecting a fifth group of components based on the average Hamming distances whereby components having an average Hamming distance under a predetermined threshold are selected.
An apparatus for processing an image including text and continuous tone regions, the apparatus comprising:
- means for performing a thresholding routine on the image to transform the image to a second image comprising components representing high contrast regions of the image;
- means for selecting a group of the components based on color variances, Hamming distances and alignment;
- means for classifying the components of the selected group according to shape to obtain classes of components;
- means for identifying components of the selected group belonging to a class of more than one component as text;
- means for identifying unidentified components of the selected group that align with components of similar color as text;
- means for identifying unidentified components of the selected group that align with the identified text as text;
- means for identifying unidentified components of the selected group that are in close proximity to the identified text and relatively small in comparison to the identified text as text;
- means for determining which components have connected black pixels to obtain a first group of components;
- means for determining which components have connected white pixels to obtain a second group of components;
- means for inverting one of the first and second groups of components to obtain a third group of components including both groups;
- means for determining a color of each component based on the image;
- means for determining a variance of color of each component;
- means for selecting a fourth group of components from the third group based on the determination of the variance of color whereby components having variance of color under a predetermined threshold are selected;
- means for calculating Hamming distances between scan lines of each component of the fourth group;
- means for calculating an average Hamming distance for each component of the fourth group;
- and,
- means for selecting a fifth group of components based on the average Hamming distances whereby components having an average Hamming distance under a predetermined threshold are selected.
Dependencies are inferred from the claim text.