Annotation free license plate recognition method and system

• US 9,536,315 B2
• Filed: 03/16/2015
• Issued: 01/03/2017
• Est. Priority Date: 01/13/2015
• Status: Active Grant

• Current Assignee: Conduent Business Services, LLC (Conduent, Inc.)
• Original Assignee: Xerox Corporation (Xerox Holdings Corp.)
• Inventors: Bulan, Orhan, Kozitsky, Vladimir, Burry, Aaron M.
• Primary Examiner(s): Liew, Alex
• Application Number:
  US14/658,713
• Publication Number:
  US 20160203380A1
• Time in Patent Office: 659 Days
• Field of Search: 382155-160, 382181-231
• US Class Current: 1/1
• CPC Class Codes: G06F 18/214 Generating training patterns; Bootstrap methods, e.g. bagging or boosting
• CPC Class Codes: G06F 18/2415 based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
• CPC Class Codes: G06V 20/63 Scene text, e.g. street names
• CPC Class Codes: G06V 30/10 Character recognition
• CPC Class Codes: G06V 30/19147 Obtaining sets of training patterns; Bootstrap methods, e.g. bagging or boosting
• CPC Class Codes: G06V 30/19173 Classification techniques

Abstract

Methods and systems for recognizing a license plate character. Synthetic license plate character images are generated for a target jurisdiction. A limited set of license plate images can be captured for a target jurisdiction utilizing an image-capturing unit. The license plate images are then segmented into license plate character images for the target jurisdiction. The license plate character images collected for the target jurisdiction can be manually labeled. A domain adaptation technique can be utilized to reduce the divergence between synthetically generated and manually labeled target jurisdiction image sets. Additionally, OCR classifiers are trained utilizing the images after the domain adaptation method has been applied. One or more input license plate character images can then be received from the target jurisdiction. Finally, the trained OCR classifier can be employed to determine the most likely labeling for the character image and a confidence associated with the label.

Claims

1. A method for recognizing a license plate character, said method comprising:

applying a domain adaptation in at least one offline training phase to reduce a divergence between synthetic license plate character images that are synthetically generated and license plate character images manually labeled for a target jurisdiction;

training in said at least one offline training phase a set of OCR classifiers using said license plate character images after said domain adaptation has been applied;

receiving in an online classification phase at least one input license plate character image from said target jurisdiction; and

determining a most likely label for a character image and a confidence associated with said label utilizing said trained OCR classifiers, thereby minimizing manual labeling required for training an OCR engine in a license plate recognition system.

2. The method of claim 1 further comprising generating in said least one offline training phase said synthetic license plate character images for said target jurisdiction.

3. The method of claim 1 further comprising manually labeling in said at least one offline training phase said license plate character images collected for said target jurisdiction.

4. The method of claim 1 further comprising:

capturing in said at least one offline training phase a limited set of license plate images for said target jurisdiction, said limited set of license plates images captured utilizing an image capturing unit.

5. The method of claim 4 further comprising:

thereafter segmenting license plate images among said limited set of license plate images into license plate character images for said target jurisdiction.

6. The method of claim 1 wherein said applying said domain adaptation in said at least one offline training phase to reduce said divergence further comprises:

reducing said divergence between a best-matched source domain and a target domain to train one-versus-all classifiers, wherein said target jurisdiction comprises said target domain and said best-matched source domain includes said synthetic license plate character images.

7. The method of claim 6 further comprising:

determining said best-matched source domain based on a similarity between a source domain and said target domain based on at least one of a distribution or a domain shift between said source domain and said target domain.

8. The method of claim 1 wherein said domain adaptation comprises an unsupervised domain adaptation if labeled samples are only available in a source domain and wherein said domain adaptation comprises a semi-supervised domain adaptation based on metric learning if some of said labeled samples are also available in a target domain, said target domain comprising said target jurisdiction.

9. The method of claim 8 wherein said image capturing unit comprises an automated license plate recognition camera.

10. A method for recognizing a license plate character, said method comprising:

in an offline training phase, generating synthetic license plate character images for a target jurisdiction;

in an offline training phase, capturing a limited set of license plate images for a target jurisdiction utilizing an image capturing unit and thereafter segmenting said license plate images into license plate character images for said target jurisdiction;

in an offline training phase, manually labeling said license plate character images collected for said target jurisdiction;

in an offline training phase, applying a domain adaptation to reduce the divergence between said synthetically generated and manually labeled target jurisdiction image sets;

in an offline training phase, training a set of OCR classifiers using said images after said domain adaptation has been applied;

in an online classification phase, receiving at least one input license plate character image from said target jurisdiction; and

utilizing said trained OCR classifiers to determine a most likely labeling for said character image and a confidence associated with said label, thereby minimizing manual labeling required for training an OCR engine in a license plate recognition system.

11. The method of claim 10 wherein said limited set of license plate images is captured utilizing an image capturing unit.

12. The method of claim 10 wherein said image capturing unit comprises an automated license plate recognition camera.

13. A system for recognizing a license plate character, said system comprising:

at least one processor; and

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

localizing a candidate region from regions of interest with respect to a tag and a tag number shown in regions of interest within a side image of a vehicle captured by said at least one camera;

applying a domain adaptation in said at least one offline training phase to reduce a divergence between a synthetic license plate character image synthetically generated and license plate character images manually labeled for a target jurisdiction;

training in said at least one offline training phase, a set of OCR classifiers using said license plate character images after said domain adaptation has been applied;

receiving in an online classification phase at least one input license plate character image from said target jurisdiction; and

determining a most likely labeling for a character image and a confidence associated with said label utilizing said trained OCR classifiers, thereby minimizing manual labeling required for training an OCR engine in a license plate recognition system.

14. The system of claim 13 wherein said instructions are further configured for generating in at least one offline training phase said synthetic license plate character images for said target jurisdiction and wherein said domain adaptation estimates a domain shift between target images of said target jurisdiction and training images and generates a shared feature representation across target samples of said target jurisdiction and said training samples.

15. The system of claim 13 wherein said instructions are further configured for manually labeling in said at least one offline training phase said license plate character images collected for said target jurisdiction.

16. The system of claim 13 wherein said instructions are further configured for capturing in said at least one offline training phase a limited set of license plate images for said target jurisdiction.

17. The system of claim 16 wherein said instructions are further configured for thereafter segmenting license plate images among said limited set of license plate images into license plate character images for said target jurisdiction.

18. The system of claim 16 wherein said limited set of license plate images is captured utilizing an image capturing unit that is operably connected to said at least one processor.

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

capturing in said at least one offline training phase a limited set of license plate images for said target jurisdiction; and

thereafter segmenting license plate images among said limited set of license plate images into license plate character images for said target jurisdiction.

20. The system of claim 19 wherein said limited set of license plate images is captured utilizing an image capturing unit.