Optical character recognition of handwritten or cursive text
AssigneeInternational Business Machines Corporation
InventorsAbdel N. Al-Karmi, Shamsher S. Singh, Baldev Singh Soor
Life of this patent
Estimating the term…
Abstract
A method for optical character recognition particularly suitable for cursive and scripted text follows the tracings of the script and encodes them as a sequence of directional vectors. Another aspect of the method adaptively preprocesses each word or sub-word of interconnected characters as a unit and the characters are accepted only when all characters in a unit have been recognized without leaving a remainder of any vectors in the unit.
Claims4
A method for optical character recognition of scanned cursive, handwritten text, comprising the steps of:
- (a) processing a scanned text image in units of interconnected characters comprising words or subwords including coding said units of interconnected characters into a sequence of directional vectors corresponding to sequential directions in said scanned text image prior to identification of single constituent characters of said words or subwords;
- (b) reducing the sequence of directional vectors to a minimum sequence to yield an abstract directional representation of intra-connected sub-units of said scanned text image, wherein said reducing step comprises the steps of:
- (i) selecting a first directional vector in said sequence;
- (ii) selecting a second directional vector in said sequence immediately following said first directional vector;
- and
- (iii) if said second directional vector is identical to said first directional vector, then discarding said second directional vector from said sequence,
- wherein said steps (i) to (iii) are repeated until no directional vector in said sequence is followed by an identical directional vector;
- (c) upon recognition of a character, inserting a marker into the sequence of vectors directly following the vector that enables recognition;
- and
- (d) moving said marker within the sequence of vectors until all vectors in the sequence contribute to recognition of the sequence of characters wherein said sequence of directional vectors are compared with grammar rules;
- and
- (e) recognizing said text based on steps (c) and (d).
An improved method for optical character recognition of handwritten cursive scanned text, comprising the steps of:
- (a) skeletonizing a scanned text image;
- (b) processing said skeletonized text image in units of interconnected characters comprising words or sub-words into a sequence of directional vectors corresponding to sequential directions in said scanned text image;
- (c) reducing the sequence of directional vectors to a minimum sequence to yield an abstract directional representation of intra-connected sub-units of said scanned text image, wherein said reducing step comprises the steps of:
- (i) selecting a first directional vector in said sequence;
- (ii) selecting a second directional vector in said sequence immediately following said first directional vector;
- and
- (iii) if said second directional vector is identical to said first directional vector, then discarding said second directional vector from said sequence,
- wherein said steps (i) to (iii) are repeated until no directional vector in said sequence is followed by an identical directional vector;
- (d) upon recognition of a character, inserting a marker into the sequence of vectors directly following the vector that enables recognition;
- (e) reparsing the sequence of directional vectors by moving the marker within the sequence of vectors until all vectors in the sequence contribute to recognition of said sequence of characters prior to identification of single constituent characters of said words or sub-words wherein said sequence of directional vectors are compared with grammar rules;
- and
- (f) recognizing said text based on steps (d) and (e).
Dependencies are inferred from the claim text.