System and methods for arabic text recognition based on effective arabic text feature extraction

• US 8,908,961 B2
• Filed: 04/23/2014
• Issued: 12/09/2014
• Est. Priority Date: 04/27/2009
• Status: Active Grant

• Current Assignee: King Abdul AZIZ City For Science and Technology
• Original Assignee: King Abdul AZIZ City For Science and Technology
• Inventors: Khorsheed, Mohammad S., Al-Omari, Hussein K.
• Primary Examiner(s): Werner, Brian P
• Application Number:
  US14/259,309
• Publication Number:
  US 20140219562A1
• Time in Patent Office: 230 Days
• Field of Search: None
• US Class Current: 382/161
• CPC Class Codes: G06V 10/50 by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
• CPC Class Codes: G06V 30/153 using recognition of characters or words
• CPC Class Codes: G06V 30/224 of printed characters having additional code marks or containing code marks
• CPC Class Codes: G06V 30/2276 with probabilistic networks, e.g. hidden Markov models
• CPC Class Codes: G06V 30/293 of characters other than Kanji, Hiragana or Katakana

Abstract

A method for automatically recognizing Arabic text includes building an Arabic corpus comprising Arabic text files written in different writing styles and ground truths corresponding to each of the Arabic text files, storing writing-style indices in association with the Arabic text files, digitizing an Arabic word to form an array of pixels, dividing the Arabic word into line images, forming a text feature vector from the line images, training a Hidden Markov Model using the Arabic text files and ground truths in the Arabic corpus in accordance with the writing-style indices, and feeding the text feature vector into a Hidden Markov Model to recognize the Arabic words.

Claims

1. A method for automatically recognizing Arabic text, comprising:

acquiring a text image comprising one or more Arabic words each including one or more Arabic characters;

identify a plurality of lines of Arabic text in the text image;

segmenting one of the plurality of lines of Arabic text into Arabic words;

digitizing at least one of the Arabic words to form a two-dimensional array of pixels each associated with a pixel value, wherein the pixel value is expressed in a binary number;

dividing the one of the Arabic words into a plurality of line images;

defining a plurality of cells in one of the plurality of line images, wherein each of the plurality of cells comprises a group of adjacent pixels;

serializing pixel values of pixels in each of the plurality of cells in one of the plurality of line images to form a binary cell number;

forming a text feature vector according to binary cell numbers obtained from the plurality of cells in one of the plurality of line images; and

feeding the text feature vector into a Hidden Markov Model to recognize the one or more Arabic words including the Arabic characters.

2. The method of claim 1, further comprising:

converting the binary cell number into a decimal cell number;

serializing the decimal cell numbers obtained from the plurality of cells in the one of the plurality of line images to form the string of decimal cell numbers; and

forming the text feature vector in accordance to a string of decimal cell numbers obtained from the plurality of cells in the one of the plurality of line images.

3. The method of claim 1, further comprising:

building an Arabic corpus comprising Arabic text files and ground truths corresponding to each of the Arabic text files, wherein the Arabic text files include Arabic texts written in different writing styles; and

storing writing-style indices in association with the Arabic text files by a computer, wherein each of the writing-style indices indicates that one of the Arabic text files is written in one of the writing styles,

wherein the Hidden Markov Model is trained using the Arabic text files and ground truths in the Arabic corpus in accordance with the writing-style indices in association with the Arabic text files.

4. The method of claim 3, wherein the writing styles specify with or without punctuation in the Arabic text.

5. The method of claim 3, wherein the writing styles specify with or without vowelization in the Arabic text.

6. The method of claim 3, wherein the writing styles specify the existence or nonexistence of a non-Arabic text in the Arabic text files.

7. The method of claim 3, wherein the step of building an Arabic corpus comprises:

automatically determining a writing style associated with one of the Arabic text files by the computer, wherein a writing-style index associated corresponding to the writing style is automatically stored in association with the one of the Arabic text files.

8. The method of claim 3, wherein the step of segmenting one of the plurality of lines of Arabic text into Arabic words comprises:

identifying two adjacent strings of Arabic characters as potential adjacent words;

comparing a spacing between the two adjacent strings of Arabic characters to a predetermined threshold; and

a word boundary is determined between two adjacent strings of Arabic characters when the spacing between the two adjacent strings of Arabic characters exceeds the predetermined threshold.

9. The method of claim 1, wherein the two-dimensional array of pixels comprises a plurality of rows in a first direction and a plurality of columns in a second direction, wherein the Arabic characters in the one of the Arabic words are aligned substantially along the first direction, wherein the plurality of line images are sequentially aligned along the first direction.

10. The method of claim 9, wherein the two-dimensional array of pixels comprises N number of rows of pixels, wherein at least one of the plurality of line images has a height defined by M number of rows in the first direction and a width defined by N number of columns in the second direction, wherein M and N are integers.

11. The method of claim 10, wherein N is in a range between 2 and about 100.

12. The method of claim 1, wherein the pixel values in the two-dimensional array of pixels are expressed in single-bit binary numbers.

13. The method of claim 1, wherein the pixel values in the two-dimensional array of pixels are expressed in multi-bit binary numbers.

14. A method for automatically recognizing Arabic text, comprising:

acquiring a text image comprising one or more Arabic words each including one or more Arabic characters;

identify a plurality of lines of Arabic text in the text image;

segmenting one of the plurality of lines of text into Arabic words;

digitizing at least one of the Arabic words to form a two-dimensional array of pixels each associated with a pixel value, wherein the pixel value is expressed in a binary number;

dividing the one of the Arabic words into a plurality of line images;

downsizing at least one of the plurality of line images to produce a downsized line image;

serializing pixel values of pixels in each column of the downsized line image to form a string of serialized numbers, wherein the string of serialized numbers forms a text feature vector; and

feeding the text feature vector into a Hidden Markov Model to recognize the one or more Arabic words including the Arabic characters.

15. The method of claim 14, wherein the two-dimensional array of pixels comprises a plurality of rows in a first direction and a plurality of columns in a second direction, wherein the Arabic characters in the one of the Arabic words are aligned substantially along the first direction, wherein the plurality of line images are sequentially aligned along the first direction.

16. The method of claim 15, wherein the two-dimensional array of pixels comprises N number of rows of pixels, wherein at least one of the plurality of line images has a height defined by M number of rows in the first direction and a width defined by N number of columns in the second direction, wherein M and N are integers.

17. The method of claim 14, further comprising:

building an Arabic corpus comprising Arabic text files and ground truths corresponding to each of the Arabic text files, wherein the Arabic text files include Arabic texts written in different writing styles; and

storing writing-style indices in association with the Arabic text files by a computer, wherein each of the writing-style indices indicates that one of the Arabic text files is written in one of the writing styles,

wherein the Hidden Markov Model is trained using the Arabic text files and ground truths in the Arabic corpus in accordance with the writing-style indices in association with the Arabic text files.

18. The method of claim 17, wherein the writing styles specify with or without punctuation in the Arabic text, with or without vowelization in the Arabic text, and the existence or nonexistence of a non-Arabic text in the Arabic text files.

19. The method of claim 14, wherein the step of segmenting one of the plurality of lines of Arabic text into Arabic words comprises:

identifying two adjacent strings of Arabic characters as potential adjacent words;

comparing a spacing between the two adjacent strings of Arabic characters to a predetermined threshold; and

a word boundary is determined between two adjacent strings of Arabic characters when the spacing between the two adjacent strings of Arabic characters exceeds the predetermined threshold.

20. The method of claim 14, wherein the step of building an Arabic corpus comprises:

automatically determining a writing style associated with one of the Arabic text files by the computer, wherein a writing-style index associated corresponding to the writing style is automatically stored in association with the one of the Arabic text files.