Two tiered text recognition

• US 7,724,957 B2
• Filed: 07/31/2006
• Issued: 05/25/2010
• Est. Priority Date: 07/31/2006
• Status: Expired due to Fees

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Abdulkader, Ahmad A.
• Primary Examiner(s): Bella; Matthew C
• Assistant Examiner(s): RAHMJOO, MANUCHEHR
• Application Number:
  US11/461,050
• Publication Number:
  US 20080025610A1
• Time in Patent Office: 1,394 Days
• Field of Search: 382/186, 382/185, 382/190, 382/195, 382/198, 382/227
• US Class Current: 382/186
• CPC Class Codes: G06V 30/10 Character recognition
• CPC Class Codes: G06V 30/226 of cursive writing
• CPC Class Codes: G06V 30/262 using context analysis, e.g. lexical, syntactic or semantic context

Abstract

Systems and methods that exploit unique properties of a language script (e.g., condition joining rules for Arabic language) to enable a two tier text recognition. In such two tier system, one tier can recognize predetermined groups of linked letters that are connected based on joining rules of a language associated with the text, and another tier dissects (and recognizes) such linked letters to respective constituent letters that form the predetermined group of linked letters. Various classifiers and artificial intelligence components can further facilitate text recognition at each level.

Claims

1. A system that facilitates text recognition comprising:

a word group component for recognition of word groups that are predefined based on joining rules of a language associated with the text, the joining rules defining a Part of an Arabic word (PAW), the word component to label the word groups based on relative horizontal overlaps between the word groups, and to extract features from the text to recognize the word groups of the text utilizing a first classifier to scale the text to a fixed size grid while maintaining an aspect ratio of the text and a second classifier based on directional codes associated with the word groups; and

a letters component for recognition of constituent letters that form the word groups, the letters component to recognize the constituent letters substantially simultaneously as the word group component to recognize the word groups.

2. The system of claim 1, wherein the letters component utilize a user-specific post processor technique to recognize the constitute letters when the constituent letters are deemed ambiguous by the letters component.

3. The system of claim 1, wherein the word group component and the letters component are Neural Net based.

4. The system of claim 1, further comprising personalization component that trains associated classifiers with allograph data.

5. The system of claim 4, wherein the allograph data comprises handwritten allograph data.

6. The system of claim 4, wherein the personalization component includes an adaptation feature to convert writer independent handwriting to personalized hand writing for a user.

7. The system of claim 6, wherein the personalization component includes a mapping feature that supplies probability for a letter appearance.

8. The system of claim 1, wherein the user-specific post processor technique utilizes a maximum likelihood linear regression (MLLR) adapted density module.

9. The system of claim 1, wherein the word group component utilizes a Beam search to recognize the word groups.

10. A computer-implementable method implemented by one or more processors executing processor-executable instructions stored in computer storage media to recognize a text, the method comprising:

simultaneously comparing, by the one or more processors, (1) a text input to predefined groups of linked letters that are connected based on joining rules for language of the input text to recognize linked letters of the text input, and (2) the text input with one or more constituent letters which form the predefined groups of linked letters;

labeling the linked letters of the text input based on relative horizontal overlaps between the linked letters; and

extracting features from the text input to recognize the linked letters of the text input utilizing a first classifier to scale the text input to a fixed size grid while maintaining an aspect ratio of the text input and a second classifier based on directional codes associated with the linked letters.

11. The method of claim 10, further comprising employing a Viterbi search, the Viterbi search including determining an edit distance between two constituent letters.

12. The method of claim 10, further comprising determining an over-segmentation scenario.

13. The method of claim 10, further comprising training a classifier based on allograph data.

14. The method of claim 13, further comprising optimizing handwriting recognition based on the training act.

15. The method of claim 14, further comprising converting a generic hand writing to personalized handwriting.

16. The method of claim 15, further comprising matching letters based on mapping of letters to a particular style.