Segmentation of text styles

In a digital processing system, a method of processing a binary text image to identify and distinguish a location of a first type style, wherein said first type style is a bold type style, and wherein said image contains at least a first region of said first type style and a second region of a second type style, comprising the step of eroding said binary text image with a fist structuring element, wherein said step of eroding comprises the steps of:

• a) thinning the image horizontally;
• b) eroding vertically using a SE having at least two adjacent vertical ON pixels;
• c) determining a ratio of a number of ON pixels in said image in a previous iteration to a number of ON pixels in a present iteration;
• d) testing said ratio by comparison to a threshold;
• and e) if said ratio is not above said threshold, returning to step a);
• to provide a first destination image, said first structuring element more likely to provide a hit in said first region than in said second region, remaining pixels used to identify said location of said first type style, and distinguish said location of said first type style from said second region of said second type style.

In a digital processing system, a method of processing a binary text image to identify and distinguish a location of a first type style, said image containing at least a first region of said first type style and a second region of a second type style, said method comprising the steps of:

• a) eroding said binary text image with a first structuring element to provide a first destination image, said first structuring element more likely to provide a hit in said first region than in said second region, remaining pixels used to identify said location of said first type style, and distinguish said location of said first type style from said second region of said second type style;
• b) using said first destination image to form a second destination image containing substantially only said first type style from said text image, comprising the steps of:
• i) dilating said first destination image to produce a third destination image;
• ii) closing said third destination image to produce a fourth destination image;
• iii) opening said fourth destination image to produce a mask;
• and
• iv) using said mask to locate said first type style in said text image to produce said second destination image;
• wherein the step of dilating is preceded by the step of reducing said first destination image.

In a digital processing system, a method of processing a binary text image to identify and distinguish a location of a first type style, said image containing at least a first region of said first type style and a second region of a second type style, comprising the step of eroding said binary text image with a first structuring element to provide a first destination image, using said first destination image to form a second destination image containing substantially only said first type style from said text image wherein the step of forming a second destination image further comprises the steps of, i) forming a seed image from said first destination image wherein said step of forming a seed image further comprises the steps of, a) dilating said first destination image vertically, b) closing said image horizontally, c) opening said image horizontally;

• d) dilating said image horizontally;
• and e) ANDing an image from step d) with said binary text image to form said seed image;
• said seed image comprising ON pixels only within regions of said first type style, ii) growing said seed image into a mask image to provide a third destination image, and iii) ANDing said third destination image with said binary text image to provide said second destination image said first structuring element more likely to provide a hit in said first region than in said second region;
• remaining pixels used to identify said location of said first type style, and distinguish said location of said first type style from said second region of said second type style.

In a digital processing system, a method of processing a binary text image to identify and distinguish a location of a first type style, said image containing at least a first region of said first type style and a second region of a second type style, comprising the step of eroding said binary text image with a first structuring element to provide a first destination image, using said first destination image to form a second destination image containing substantially only said first type style from said text image wherein the step of forming a second destination image further comprises the steps of, i) forming a seed image from said first destination image;

• said seed image comprising ON pixels only within regions of said first type style, ii) growing said seed image into a mask image to provide a third destination image, wherein said step of growing said seed image into a mask image further comprises the steps of:
• a) saving said seed image as a saved image;
• b) dilating said seed image;
• c) ANDing an image from step b) with said mask image to produce a partially filled seed image;
• d) comparing said partially filled seed image with said saved image;
• e) if said seed image has not changed from a previous iteration, outputting said partially filled seed image as said third destination image;
• and f) if said seed image differs from a previous iteration, saving said partially filled seed as a saved image and returning to step b);
• and iii) ANDing said third destination image with said binary text image to provide said second destination image said first structuring element more likely to provide a hit in said first region than in said second region;
• remaining pixels used to identify said location of said first type style, and distinguish said location of said first type style from said second region of said second type style.

An optical character recognition system for identifying characters of a first type style in a document, said document containing characters of a first type style and a second type style, comprising:

• a) input means for inputting a binary text image of said document;
• b) means for identifying a region of said first type style programmed to erode said input binary text image with an structural element to provide a first destination image having a plurality of remaining pixels, said structural element more likely to provide a hit in said region of said first type style than in other regions, and said region identifying means programmed to create a mask image from said plurality of remaining pixels of said first destination image, said mask image substantially covering characters of said first type style;
• and
• c) means for identifying characters substantially covered by said mask image.

In a digital processing means, a method of identifying handwritten annotation areas of an input image having handwritten annotation areas and machine printed text areas, said method comprising the steps of:

• a) morphologically processing a region of said input image having a plurality of handwritten and machine printed characters to produce an image having a plurality of remaining pixels concentrated in either said machine printed text areas or said handwritten annotation areas;
• b) morphologically processing said plurality of remaining pixels to produce a mask image substantially masking either said machine printed text areas or said handwritten annotation areas;
• and
• c) using said mask image to produce a destination image, said destination image identifying substantially only said machine printed text or said handwritten annotations areas.