Back to results
US 5,182,777 AGranted
Method for extracting feature quantities of a character
AssigneeRicoh Company, Ltd.
InventorsHiroshi Nakayama, Keiji Kojima, Gen Sato
Life of this patent
Estimating the term…
Abstract
In a method of extracting feature quantities of a character, a pair of region dividing lines which intersect a character image are obtained by using generated projections of direction codes. Then, a reference position for extracting feature quantities of the character is decided by finding positions at which the region dividing lines intersect the character image. Thereafter, the feature quantities of the character having the reference position are extracted by subjecting the codes of the divided segments to a predetermined operation for obtaining information regarding a shape of each of the divided segments.
Claims20
A method of extracting feature quantities of a character including the steps of:
- optically scanning a character to be recognized to thereby generate a binary-valued character image;
- extracting feature information from a contour pattern of said binary-valued character image;
- and
- extracting feature quantities of said character from said feature information by using information on dividing lines for dividing said binary-valued character image into a plurality of portions, characterized by comprising:
- tracing said contour portion of said binary-valued character image to thereby generate feature information on said contour portion;
- setting a first region dividing line at a first position within the width of said contour portion viewed from an end of a circumscribed rectangular region to which said binary-valued character image is circumscribed;
- setting a second region dividing line at a second position within the width of said contour portion viewed from an opposite end of said circumscribed rectangular region, said first and said second region dividing lines being used for extracting feature quantities of said character;
- detecting intersection position where each of said first and second region dividing lines intersects said contour portion;
- counting the number of said intersecting positions;
- and
- extracting feature quantities of said character from said feature information on said contour portion by using information on said intersecting positions, said feature quantities being represented by said counted number of said intersecting positions whereby said feature quantities are used to recognize said character.
8 more elements
The method of extracting feature quantities of a character including the steps of:
- optically scanning a character to be recognized to thereby generate a binary-valued character image;
- extracting feature information from a contour pattern of said binary-valued character image;
- extracting feature quantities of said character from said feature information by using information on dividing lines for dividing said binary-valued character image into a plurality of portions, characterized by comprising:
- tracing said contour portion of said binary-valued character image to thereby generate feature information on said contour portion;
- setting a first region dividing line at a first position within the width of said contour portion viewed from an end of a circumscribed rectangular region to which said binary-valued character image is circumscribed;
- setting a second region dividing line at a second position within the width of said contour portion viewed from an opposite end of said circumscribed rectangular region, said first and said second region dividing lines being used for extracting feature quantities of said character;
- detecting intersection position where each of said first and second region dividing lines intersects said contour portion;
- dividing said contour portion into a plurality of segments at said intersecting points so that said feature information on said contour portion is divided into partitioned feature information sequences relating to the corresponding segments;
- and
- extracting feature quantities of said character from said feature information on said contour portion by using information on said intersecting positions, from each of said partitioned feature information sequences whereby said feature quantities are used to recognize said character.
7 more elements
A method of extracting feature quantities of a character including the steps of:
- optically scanning a character to be recognized to thereby generate a binary-valued character image;
- extracting feature information from a contour pattern of said binary-valued character image;
- extracting feature quantities of said character from said feature information by using information on dividing lines for dividing said binary-valued character image into a plurality of portions, characterized by comprising:
- tracing said contour portion of said binary-valued character image to thereby generate feature information on said contour portion;
- setting a first region dividing line at a first position within the width of said contour portion viewed from an end of a circumscribed rectangular region to which said binary-valued character image is circumscribed;
- setting a second region dividing line at a second position within the width of said contour portion viewed from an opposite end of said circumscribed rectangular region, said first and said second region dividing lines being used for extracting feature quantities of said character;
- detecting intersection position where each of said first and second region dividing lines intersects said contour portion;
- equally dividing said feature information on said contour portion into a plurality of partitioned feature information sequences with respect to a reference position which is one of said intersecting positions;
- and
- extracting feature quantities of said character from said feature information on said contour portion by using information on said intersecting positions, from each of said partitioned feature information sequences whereby said feature quantities are used to recognize said character.
7 more elements
Dependencies are inferred from the claim text.