REINFORCEMENT LEARNING APPROACH TO CHARACTER LEVEL SEGMENTATION OF LICENSE PLATE IMAGES

• US 20150125041A1
• Filed: 01/21/2014
• Published: 05/07/2015
• Est. Priority Date: 11/06/2013
• Status: Active Grant

• Current Assignee: Conduent Business Services, LLC (Conduent, Inc.)
• Original Assignee: Xerox Corporation (Xerox Holdings Corp.)
• Inventors: Burry, Aaron Michael, Abtahi, Farnaz
• Granted Patent: US 9,213,910 B2
• Time in Patent Office: Days
• US Class Current: 382/105
• CPC Class Codes: G06F 18/214 Generating training patterns; Bootstrap methods, e.g. bagging or boosting
• CPC Class Codes: G06V 10/26 Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
• CPC Class Codes: G06V 20/56 exterior to a vehicle by using sensors mounted on the vehicle
• CPC Class Codes: G06V 20/62 Text, e.g. of license plates, overlay texts or captions on TV images
• CPC Class Codes: G06V 20/625 License plates

Abstract

Methods and systems for achieving accurate segmentation of characters with respect to a license plate image utilizing a reinforcement learning approach. A vehicle image can be captured by an image capturing unit and processed utilizing an ALPR (Automatic License Plate Recognition) unit. The reinforcement learning (RL) approach can be configured to initialize a segmentation agent with a starting location. A proper segmentation path (cuts) from top to bottom and from a darker to lighter area in a cropped license plate image can be identified by the segmentation agent during a training phase. Rewards can be provided based on a number of good and bad moves. The association between a current state and a sensory input with a preferred action can be learned by the segmentation agent at the end of the training phase.

Claims

1. A license plate character segmentation method, said method comprising:

initializing a segmentation agent in association with a starting location to identify a segmentation path in a license plate image by said segmentation agent during a training phase;

automatically learning an association between a current state and a sensory input with a preferred action by said segmentation agent at an end of said training phase; and

traversing from a top to a bottom of said license plate image by said segmentation agent at runtime utilizing said association learned during said training phase to improve a robustness of license plate character segmentation and an operational latitude of an automatic license plate recognition unit.

2. The method of claim 1 further comprising executing said segmentation path from a particular point on a top border of a text area associated with said license plate image to a bottom border thereof and from a darker area to a lighter area without entering a body of characters associated with said license plate image.

3. The method of claim 1 further comprising:

employing a pixel neighborhood of size N×

N as a field of view; and

comparing a mean gray value of said neighborhood as seen in a potential move direction to that of a neighborhood centered on a current pixel location.

4. The method of claim 1 further comprising providing a reward to said segmentation agent during a training phase based on a number of good moves and bad moves.

5. The method of claim 4 further comprising:

evaluating said segmentation move from a current pixel location to a next pixel location;

designating a move from a light region to a dark region and/or an upward move with respect to said license plate image as a bad move; and

designating a move from a dark region to a light region and/or a downward move with respect to said license plate image as a good move.

6. The method of claim 1 further comprising determining said segmentation path beginning with every possible pixel location in a top row of said license plate image by said segmentation agent.

7. The method of claim 1 further comprising a first segmentation to determine initial locations for candidate segmentation paths to be evaluated by said segmentation agent.

8. The method of claim 7 wherein said segmentation agent determines a validity of segmentation paths produced by said first segmentation.

9. The method of claim 7 wherein said first segmentation comprises a projective segmentation with an appropriate threshold selected.

10. A license plate character segmentation system, said system comprising:

a processor; and

a computer-usable medium embodying computer program code, said computer-usable medium capable of communicating with said processor, said computer program code comprising instructions executable by said processor and configured for;

initializing a segmentation agent in association with a starting location to identify a segmentation path in a license plate image by said segmentation agent during a training phase;

automatically learning an association between a current state and a sensory input with a preferred action by said segmentation agent at an end of said training phase; and

traversing from a top to a bottom of said license plate image by said segmentation agent at runtime utilizing said association learned during said training phase to improve a robustness of license plate character segmentation and an operational latitude of an automatic license plate recognition unit.

11. The system of claim 10 wherein said instructions are further configured for executing said segmentation path from a particular point on a top border of a text area associated with said license plate image to a bottom border thereof and from a darker area to a lighter area without entering a body of characters associated with said license plate image.

12. The system of claim 10 wherein said instructions are further configured for:

employing a pixel neighborhood of size N×

N as a field of view; and

comparing a mean gray value of said neighborhood as seen in a potential move dire ion to that of a neighborhood centered on a current pixel location.

13. The system of claim 10 wherein said instructions are further configured for providing a reward to said segmentation agent during a training phase based on a number of good moves and bad moves.

14. The system of claim 13 wherein said instructions are further configured for:

evaluating said segmentation move from a current pixel location to a next pixel location;

designating a move from a light region to a dark region and/or an upward move with respect to said license plate image as a bad move; and

designating a move from a dark region to a light region and/or a downward move with respect to said license plate image as a good move.

15. The system of claim 10 wherein said instructions are further configured for determining said segmentation path beginning with every possible pixel location in a top row of said license plate image by said segmentation agent.

16. The system of claim 10 wherein said instructions are further configured for comprising a first segmentation to determine initial locations for candidate segmentation paths to be evaluated by said segmentation agent.

17. The method of claim 16 wherein said segmentation agent determines a validity of segmentation paths produced by said first segmentation.

18. The method of claim 16 wherein said first segmentation comprises a projective segmentation with an appropriate threshold selected.

19. A processor-readable medium storing code representing instructions to cause a process for character segmentation, said code comprising code to:

initialize a segmentation agent in association with a starting location to identify a segmentation path in a license plate image by said segmentation agent during a training phase;

automatically learn an association between a current state and a sensory input with a preferred action by said segmentation agent at an end of said training phase; and

traverse from a top to a bottom of said license plate image by said segmentation agent at runtime utilizing said association learned during said training phase to improve a robustness of license plate character segmentation and an operational latitude of an automatic license plate recognition unit.

20. The processor-readable medium of claim 19 wherein said code further comprises code to:

execute said segmentation path from a particular point on a top border of a text area associated with said license plate image to a bottom border thereof and from a darker area to a lighter area without entering a body of characters associated with said license plate image;

employ a pixel neighborhood of size N×

N as a field of view; and

compare a mean gray value of said neighborhood as seen in a potential move direction to that of a neighborhood centered on a current pixel location.