ONLINE ARABIC HANDWRITING RECOGNITION

• US 20100128985A1
• Filed: 07/26/2007
• Published: 05/27/2010
• Est. Priority Date: 07/27/2006
• Status: Active Grant

• Current Assignee: BGN Technologies Ltd.
• Original Assignee: BGN Technologies Ltd.
• Inventors: Biadsy, Fadi, El-Sana, Jihad
• Granted Patent: US 8,615,131 B2
• Time in Patent Office: Days
• US Class Current: 382/189
• CPC Class Codes: G06V 30/32 Digital ink

Abstract

Method for online character recognition of Arabic text, the method including receiving handwritten Arabic text from a user in the form of handwriting strokes, sampling the handwriting strokes to acquire a sequence of two dimensional point representations thereof, with associated temporal data, geometrically pre processing and extracting features on the point representations, detecting delayed strokes and word parts in the pre processed point representations, projecting the delayed strokes onto the body of the word parts, constructing feature vector representations for each word part, thereby generating an observation sequence, and determining the word with maximum probability given the observation sequence, resulting in a list of word probabilities.

Claims

1. A method for online character recognition of Arabic text, the method comprising the procedures of:

receiving handwritten Arabic text from a user in the form of handwriting strokes;

sampling said handwriting strokes to acquire a sequence of two-dimensional point representations thereof, including associated temporal data;

geometrically pre-processing and extracting features on said point representations;

detecting delayed strokes and word-parts in said pre-processed point representations;

projecting said delayed strokes onto the body of said word-parts;

constructing feature vector representations for each said word-part, thereby generating an observation sequence; and

determining the word with maximum probability given said observation sequence, resulting in a list of word probabilities.

2. The method according to claim 1, further including the procedure of providing at least one word having the highest resultant word probability, to be user selectable when more than one word is provided.

3. The method according to claim 1, further including the procedure of developing a respective probabilistic model for each letter-shape, based on a series of received predetermined words.

4. The method according to claim 3, wherein said procedure of developing a respective probabilistic model for each letter-shape includes the sub-procedures of:

online receiving a series of predetermined handwritten Arabic words from at least one trainer;

identifying letter-shapes in said received words;

generating a plurality of Hidden Markov Models (HMMs), each of said HMMs representing a letter-shape;

generating word dictionaries indexed in accordance with word features; and

establishing word-part networks and associating said HMMs therewith.

5. The method according to claim 1, wherein said word with maximum probability is determined by performing a Viterbi algorithm.

6. The method according to claim 1, wherein said features are selected from the list consisting of:

the number of word-parts in a word;

the local-angle;

the super-segment;

the presence of loops;

the position of a loop in a word-part;

the presence of cusps;

the presence of crossings;

the number of closed curves;

the number of indented sections;

the number of dots;

the location of a dot;

the type of a delayed stroke;

an indication whether a word-part begins with a curved shape; and

the context of a word.

7. The method according to claim 1, wherein said procedure of geometrically pre-processing includes a type of processing selected from the list consisting of:

reducing noise and imperfections in said point representations;

eliminating redundant point samples;

reducing the sampling rate;

interpolation of said point representations;

increasing the sampling rate; and

re-sampling said point representations

8. The method according to claim 1, wherein said procedure of detecting delayed strokes and word-parts includes:

generating boundary lines around a word-part;

searching for markings within the region encompassed by said boundary lines;

detecting markings within said region; and

determining if each of said markings is a delayed stroke associated with said word-part.

9. The method according to claim 1, wherein said procedure of projecting said delayed strokes onto the body of said word-parts includes, for each of said delayed strokes associated with a word-part,

creating a first vector extending from the first point of said delayed stroke point sequence to a point on the body of said word-part point sequence;

creating a second vector extending from the last point of said delayed stroke point sequence to a next point on the body of said word-part point sequence; and

creating a new point sequence that includes said delayed stroke point sequence, said word-part point sequence, said first vector point sequence, and said second vector point sequence.

10. The method according to claim 1, wherein said procedure of determining the word with maximum probability includes limiting the word domain in accordance with the type of user.

11. The method according to claim 4, further including the procedure of geometrically pre-processing and extracting features on said letter-shapes.

12. A system for online character recognition of Arabic text, the system comprising:

a user handwriting strokes interface, receiving handwritten Arabic text from a user in the form of handwriting strokes, and sampling said handwriting strokes to acquire a sequence of two-dimensional point representations thereof, including associated temporal data;

a character recognition processor, coupled with said user handwriting strokes interface, said character recognition processor geometrically pre-processing and extracting features on said point representations, detecting delayed strokes and word-parts in said pre-processed point representations, projecting said delayed strokes onto the body of said word-parts, constructing feature vector representations for each said word-part, thereby generating an observation sequence, and determining the word with maximum probability given said observation sequence, resulting in a list of word probabilities; and

a storage unit, coupled with said user handwriting strokes interface and with said character recognition processor.

13. The system according to claim 12, wherein said system operates on a graphics tablet.

14. The system according to claim 12, wherein said system operates on a computer coupled with a graphics tablet.

15. The system according to claim 12, wherein said user writes said handwritten Arabic text with a digital writing instrument.

16. The system according to claim 13, wherein said graphics tablet includes an input screen selected from the list consisting of:

pressure sensitive;

touch sensitive; and

light sensitive.