Methods for determining font attributes of characters
AssigneeXerox Corporation
InventorsZhigang Fan, Robert Cooperman, Robert Shuchatowitz, Lucy Hadden, Emil Rainero, Frederick Roberts, Jr.
Life of this patent
Estimating the term…
Abstract
Font attributes of characters or letters are estimated by an optical character recognition system using character, run or stroke-based techniques with or without character identification information.
Claims24
A method for estimating at least one stroke-based font attribute of a group of characters of an input image independently of any character identification information of the group of characters, comprising:
- decomposing each character of the group of characters into strokes;
- generating, for each stroke, two best fit parabolas, each best fit parabola having parameters;
- classifying each stroke based on the parameters from the two best fit parabolas corresponding to that stroke, each best fit parabola having a curvature parameter, an orientation parameter and a position parameter;
- forming at least one histogram from the classified strokes;
- identifying at least one frequency peaks of the at least one stroke histogram;
- and
- estimating at least one stroke-based attribute based on the at least one identified frequency peak.
A method for estimating at least one stroke-based font attribute of a group of characters of an input image independently of any character identification information of the group of characters, comprising:
- decomposing each character of the group of characters into strokes;
- defining a bounding box around each character of the group of characters;
- identifying vertical strokes from said strokes
- determining, for each vertical stroke, an end width that touches one of a top and a bottom of the bounding box defined around the character of which that stroke is a portion;
- estimating serifness of the group of characters based on the determined end widths of the vertical strokes touching the top or bottom of the corresponding bounding box.
Dependencies are inferred from the claim text.