Systems and associated methods for arabic handwriting using synthetic extension concatenation and synthesis

Patent metadata

Field	Value
Publication number	US-10055659-B2
Application number	US-201615145582-A
Country	US
Kind code	B2
Filing date	May 3, 2016
Priority date	May 4, 2015
Publication date	Aug 21, 2018
Grant date	Aug 21, 2018

Abstract

Systems and associated methodology are presented for Arabic handwriting synthesis including accessing character shape images of an alphabet, determining a connection point location between two or more character shapes based on a calculated right edge position and a calculated left edge position of the character shape images, extracting character features that describe language attributes and width attributes of characters of the character shape images, the language attributes including character Kashida attributes, and generating images of cursive text based on the character Kashida attribues and the width attribues.

First claim

The invention claimed is:

1. A system for handwriting synthesis comprising: circuitry configured to:
access character shape images of an alphabet,
determine a connection point location between two or more character shapes based on a calculated right edge position and a calculated left edge position of the character shape images,
extract character features that indicate language attributes and width attributes of characters of the character shape images, the language attributes including character Kashida attributes,
identify Kashida extensions as part of the character Kashida attributes;
isolate the identified Kashida extensions from pepper noise components based on a predetermined ground-truth label, by constraining the extracted character features to be two consecutive characters,
extract the identified Kashida extensions based on the predetermined ground-truth label, and
generate images of cursive text based on the character Kashida attributes and the width attributes.

2. The system of claim 1 , wherein the circuitry is further configured to remove a left edge segment and a right edge segment from the identified Kashida extensions.

3. The system of claim 2 , wherein a width of each of the left edge segment and the right edge segment is adaptively computed based on a Kashida width based on the calculated right edge position and the calculated left edge position.

4. The system of claim 1 , wherein each extracted Kashida is classified based on at least: a width of the extracted Kashida, a slope of an upper contour direction (UCD) of the extracted Kashida, and a slope of a lower contour direction (LCD) of the extracted Kashida.

5. The system of claim 4 , wherein the circuitry is further configured to:
generate a width probability density function (PDF) for each of the width, UCD and LCD of the extracted Kashida,
wherein the width PDF is generated based on one or more selected square bins having a width of 8-pixels, and
discard extracted Kashida having a width of less than 6-pixels.

6. The system of claim 5 , wherein the width PDF is further generated based on at least one of an author related attribute of the character shape, a character from which the extracted Kashida originates, or a character to which the extracted Kashida connects.

7. The system of claim 1 , wherein the circuitry is further configured to filter out attributes relating to a thickness of a left edge segment and a thickness of a right edge segment of the extracted Kashida.

8. A method for outputting synthesized handwritten text comprising:
accessing, with circuitry, character shape images of an alphabet;
determining, with the circuitry, a connection point location between two or more character shapes based on a calculated right edge position and a calculated left edge position of the character shape images;
extracting, with the circuitry, character features that indicate language attributes and width attributes of characters of the character shape images, the language attributes including character Kashida attributes;
identifying, with the circuitry, Kashida extensions as part of the character Kashida attributes;
isolating, with the circuitry, the identified Kashida extensions from pepper noise components based on a predetermined ground-truth label, by constraining the extracted character features to be two consecutive characters;
extracting, with the circuitry, the identified Kashida extensions based on the predetermined ground-truth label; and
generating, with the circuitry, images of cursive text based on the character Kashida attributes and the width attributes.

9. The method of claim 8 , further comprising
removing a left edge segment and a right edge segment from the identified Kashida extensions.

10. The method of claim 9 , wherein a width of each of the left edge segment and the right edge segment is adaptively computed based on a Kashida width based on the calculated right edge position and the calculated left edge position.

11. The method of claim 9 , further comprising:
classifying each extracted Kashida based on at least: a width of the extracted Kashida, a slope of an upper contour direction (UCD) of the extracted Kashida, and a slope of a lower contour direction (LCD) of the extracted Kashida.

12. The method of claim 11 , further comprising:
generating a width probability density function (PDF) for each of the width, UCD and LCD of the extracted Kashida,
wherein the width PDF is generated based on one or more selected square bins having a width of 8-pixel; and
discarding extracted Kashida having a width of less than 6-pixels.

13. The method of claim 12 , further comprising:
generating the width PDF based on at least one of an author related attribute of the character shape, a character from which the extracted Kashida originates, or a character to which the extracted Kashida connects.

14. The method of claim 8 , further comprising:
filtering out attributes relating to a thickness of a left edge segment and a thickness of a right edge segment of the extracted Kashida.

15. A non-transitory computer readable medium having computer-readable instructions thereon which when executed by a computer cause the computer to perform a method, the method comprising:
accessing character shape images of an alphabet;
determining a connection point location between two or more character shapes based on a calculated right edge position and a calculated left edge position of the character shape images;
extracting character features that indicate language attributes and width attributes of characters of the character shape images, the language attributes including character Kashida attributes;
identifying Kashida extensions as part of the character Kashida attributes;
isolating the identified Kashida extensions from pepper noise components based on a predetermined ground-truth label, by constraining the extracted character features to be two consecutive characters;
extracting the identified Kashida extensions based on the predetermined ground-truth label; and
generating images of cursive text based on the character Kashida attributes and the width attributes.

Assignees

• Univ King Fahd Pet & Minerals

Inventors

• Elarian Yousef S I

Classifications

• G06V30/2264

  using word shape · CPC title

• G06F21/36 Primary

  by graphic or iconic representation · CPC title

• G06F18/2321

  using statistics or function optimisation, e.g. modelling of probability density functions · CPC title

• G06V10/768

  using context analysis, e.g. recognition aided by known co-occurring patterns · CPC title

• G06K9/4604

  Physics · mapped topic