Method and system for identifying a license plate

• US 8,792,682 B2
• Filed: 04/21/2011
• Issued: 07/29/2014
• Est. Priority Date: 04/21/2011
• Status: Active Grant

• Current Assignee: Conduent Business Services, LLC (Conduent, Inc.)
• Original Assignee: Xerox Corporation (Xerox Holdings Corp.)
• Inventors: Fan, Zhigang, Kozitsky, Vladimir, Burry, Aaron M.
• Primary Examiner(s): Lu, Tom Y
• Assistant Examiner(s): YANG, WEI WEN
• Application Number:
  US13/091,371
• Publication Number:
  US 20120269398A1
• Time in Patent Office: 1,195 Days
• Field of Search: 382/105, 382/100, 382/148, 382/154
• US Class Current: 382/105
• CPC Class Codes: G06V 20/625 License plates
• CPC Class Codes: G06V 20/63 Scene text, e.g. street names
• CPC Class Codes: G06V 30/10 Character recognition
• CPC Class Codes: G06V 30/184 by analysing segments intersecting the pattern

Abstract

A license plate localization method and system based on a combination of a top-down texture analysis and a bottom-up connected component. An image with respect to a vehicle captured by an image capturing unit can be processed in order to locate and binarize a busy area. A black run with respect to the binarized image can be analyzed and classified and one or more objects (connected components) can be generated based on the black run classification. The objects can be further classified in accordance with their size utilizing a run-length based filter to filter out a non-text object. The leftover objects can then be spatially clustered and the uniformity and linearity of the clustered objects can be examined based on a linearity test. The clustered objects can be rejected if they fail the linearity test and the detected objects can further be matched with a plate edge characteristic in order to locate a license plate.

Claims

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

analyzing an image of a vehicle utilizing a top-down analysis method to identify a potential license plate area of a license plate by producing a binarized image of said image of said vehicle;

locating and binarizing a busy area of said image of said vehicle and thereafter classifying a black run with respect to a binarized image and generating a plurality of objects based on a black run classification;

classifying said black run of said binarized image into at least three categories based on length by a run classification unit;

utilizing a run-length based filter to filter out non-text objects by retaining at least two smallest categories of said at least three categories of said black run in order to thereafter spatially cluster at least one leftover object;

analyzing said at least one leftover object utilizing a bottom-up analysis method by examining a uniformity and a linearity of a clustered object of said at least one leftover object based on a linearity test in order to thereafter match a detected object with a plate edge characteristic in order to locate a license plate; and

generating a license plate identification of said license plate.

2. The method of claim 1 further comprising rejecting said clustered object if said clustered object fails said linearity test.

3. The method of claim 1 further comprising high-pass filtering said image of said vehicle prior to a binarization of said image of said vehicle in order to adapt to a plurality of background colors for use in locating and identifying said license plate.

4. The method of claim 1 wherein said black run comprises at least one of the following types of runs:

a tiny run;

a short run;

a median run; and

a long run.

5. The method of claim 1 wherein said run-length based filter further comprises:

setting at least two thresholds that correspond to a maximum vertical stroke width and a maximum run-length;

classifying a run as a tiny run if said length is smaller than said maximum vertical stroke width and as a short run if said length is between said maximum vertical stroke width and said maximum run-length; and

setting a third threshold to separate remaining runs into a median run and a long run in order to thereafter delete said median run and said long run that are not a part of a text to reduce an opportunity of a text-frame touching.

6. The method of claim 1 wherein said run-length based filter further comprises:

re-classifying a run that is initially classified as a tiny run or a short run if said run is adjacent to a long run; and

discarding long neighboring runs and performing a connected component creation by identifying said tiny run as a start point wherein said plurality of objects grow recursively by adding said tiny run and said short run that touches said plurality of objects.

7. The method of claim 1 wherein said linearity test further comprises:

cropping an area slightly larger than a bounding box of a said cluster from an image of a vehicle if said cluster is detected in order to estimate a dominant orientation for said area slightly larger than a bounding box;

determining an orientation of a top and bottom line formed by a text if said license plate does not have a visible boundary in order to thereafter project said clustered object to a direction detected to determine whether a top pixel and a bottom pixel form at least two lines; and

calculating a fitting error as a result of said linearity test.

8. The method of claim 1 further comprising transmitting said identified license plate to a module in order to thereafter receive feedback from said module to alter a detection decision.

9. The method of claim 8 further comprising at least one of the following types of module:

a character segmentation module;

an optical character recognition module; and

/or a license plate matching module.

10. A system for locating a license plate, said system comprising:

a processor;

a data bus coupled to said processor;

a computer-usable medium embodying computer code, said computer-usable medium being coupled to said data bus, said computer program code comprising instructions executable by said processor and configured for;

analyzing an image of a vehicle utilizing a top-down analysis method to identify a potential license plate area of a license plate by producing a binarized image of said image of said vehicle;

locating and binarizing a busy area of said image of said vehicle and thereafter classifying a black run with respect to a binarized image and generating a plurality of objects based on a black run classification;

classifying said black run of said binarized image into at least three categories based on length by a run classification unit;

utilizing a run-length based filter to filter out non-text objects by retaining at least two smallest of said at least three categories of said black run in order to thereafter spatially cluster at least one leftover object;

analyzing said at least one leftover object utilizing a bottom-up analysis method by examining a uniformity and a linearity of a clustered object of said at least one leftover object based on a linearity test in order to thereafter match a detected object with a plate edge characteristic in order to locate a license plate; and

generating a license plate identification of said license plate.

11. The system of claim 10 wherein said instructions are further configured for rejecting said clustered object if said clustered object fails said linearity test.

12. The system of claim 10 wherein said instructions are further configured for high-pass filtering said image of said vehicle prior to said binarization of said image of said vehicle in order to adapt to a plurality of background colors for use in locating and identifying said license plate.

13. The system of claim 10 wherein said run-length based filter further comprises:

setting at least two thresholds that correspond to a maximum vertical stroke width and a maximum run-length;

classifying a run as a tiny run if said length is smaller than said maximum vertical stroke width and as a short run if said length is between said maximum vertical stroke width and said maximum run-length; and

setting a third threshold to separate remaining runs into a median run and a long run in order to thereafter delete said median run and said long run that are not a part of a text to reduce an opportunity of a text-frame touching.

14. The system of claim 10 wherein said run-length based filter further comprises:

re-classifying a run that is initially classified as a tiny run or a short run if said run is adjacent to a long run; and

discarding long neighboring runs and performing a connected component creation by identifying said tiny run as a start point wherein said plurality of objects grow recursively by adding said tiny run or said short run that touches said plurality of objects.

15. The system of claim 10 wherein said linearity test further comprises:

cropping an area slightly larger than a bounding box of a cluster from said image of said vehicle if said cluster is detected in order to estimate a dominant orientation for said area slightly larger than a bounding box;

determining an orientation of a top and bottom line formed by a text if said license plate does not have a visible boundary in order to thereafter project said object in said cluster to a direction detected to determine whether a top pixel and a bottom pixel form at least two lines; and

calculating a fitting error as a result of said linearity test.

16. A processor-readable non-transitory storage medium storing code representing instructions to cause a processor to perform a process to locate a license plate, said code comprising code to:

analyze an image of a vehicle utilizing a top-down analysis method to identify a potential license plate area of a license plate by producing a binarized image of said image of said vehicle;

locate and binarize a busy area of said image of said vehicle and thereafter classify a black run with respect to a binarized image and generate a plurality of objects based on a black run classification;

classify said black run of said binarized image into at least three categories based on length by a run classification unit;

utilize a run-length based filter to filter out non-text objects by retaining at least two smallest of said at least three categories of said black run in order to thereafter spatially cluster at least one leftover object;

analyze said at least one leftover object utilizing a bottom-up analysis method by examining a uniformity and a linearity of a clustered object of said at least one leftover object based on a linearity test in order to thereafter match a detected object with a plate edge characteristic in order to locate a license plate; and

generate a license plate identification of said license plate.