Select type of test image based on similarity score to database image

• US 9,773,194 B2
• Filed: 01/28/2016
• Issued: 09/26/2017
• Est. Priority Date: 01/28/2016
• Status: Active Grant

• Current Assignee: Longsand Limited (Open Text Corporation)
• Original Assignee: Longsand Limited (Open Text Corporation)
• Inventors: Rosoman, Tom, Bettinson, David, Smith, Chris, Ayo Aresti, Unai
• Primary Examiner(s): Werner, Brian P
• Application Number:
  US15/009,156
• Publication Number:
  US 20170220898A1
• Time in Patent Office: 607 Days
• CPC Class Codes: G06V 20/10 Terrestrial scenes scenes under surveillance with static cameras G06V20/52; scenes perceived from the exterior of a vehicle G06V20/56; scenes perceived from the interior of a vehicle G06V20/59
• CPC Class Codes: G06V 20/625 License plates
• CPC Class Codes: G06V 20/63 Scene text, e.g. street names
• CPC Class Codes: G06V 2201/09 Recognition of logos

Abstract

A homography is determined between a test image and a plurality of database images based on a location of a character along a main line of text of the test image. A similarity score is determined for each of the plurality of database images based on the determined homography. The similarity score may measure a similarity of a characteristic between the test image and the corresponding database image. The characteristic may be at a location separate from the main line of text of the test image. A type of the test image may be selected based on the similarity scores.

Claims

1. A device, comprising:

at least one processor, and

a memory storing instructions executable by the at least one processor to;

determine a homography between a test image and a plurality of database images based on a location of a character along a main line of text of the test image;

determine a similarity score for each of the plurality of database images based on the determined homography, the similarity score to measure a similarity of a characteristic between the test image and the corresponding database image, the characteristic to be at a location separate from the main line of text of the test image; and

select a type of the test image based on the similarity scores.

2. The device of claim 1, wherein,

the instructions to determine the similarity score further comprise instructions to determine a similarity set of similarity scores for each of the plurality of database images, and

each of the similarity scores of the similarity set is to relate to a different characteristic.

3. The device of claim 2, wherein,

the similarity scores of the similarity set are to include at least one of a sub-read score, logo score and color score,

the sub-read score is to relate to an accuracy in identifying text at a location separate from the main line of text of the test image,

the logo score is to relate to an accuracy in identifying a logo at a location separate from the main line of text of the test image, and

the color score is to relate to an accuracy in identifying a color at a location of the test image.

4. The device of claim 3, wherein,

the similarity set of a first database image of the plurality of database images is to include the sub-read score, if the first database image includes a sub-read,

the instructions to determine the similarity score further comprise instructions to map a location of the sub-read of the first database image to a corresponding location of the test image based on the homography, and

the instructions to determine the similarity score further comprise instructions to calculate the sub-read score for the similarity set of the first database image based on a confidence that a character at the corresponding location of the test image is identified.

5. The device of claim 4, wherein,

the instructions to determine the similarity score further comprise instructions to calculate the sub-read score for a remainder of the plurality of database images that include a sub-read, and

the instructions to determine the homography further comprise instructions to refine the homography based on the location of the sub-read of the test image, if the test image is determined to include the sub-read.

6. The device of claim 3, wherein,

the similarity set of a first database image of the plurality of database images is to include the logo score, if the first database image includes a logo,

the instructions to determine the similarity score further comprise instructions to de-skew a location of the test image corresponding to a location of a logo in the first database image based on the homography, and

the instructions to determine the similarity score further comprise instructions to calculate the logo score for the similarity set of the first database image based on a confidence that a logo at the corresponding location of the test image matches the logo of the first database image.

7. The device of claim 6, wherein

the instructions to determine the similarity score further comprise instructions to calculate the logo score for a remainder of the plurality of database images that include a logo, and

the instructions to determine the homography further comprise instructions to refine the homography based on the location of the logo of the test image, if the test image is determined to include the logo.

8. The device of claim 3, wherein

the similarity set of a first database image of the plurality of database images is to include the color score, if the first database image includes a color,

the instructions to determine the similarity score further comprise instructions to map a region of the first database image having the color to a corresponding region of the test image based on the homography, and

the instructions to determine the similarity score further comprise instructions to calculate the color score for the similarity set of the first database image based on a percentage of the regions of the first database image and the test image that have a same color.

9. The device of claim 8, wherein,

the instructions to determine the similarity score further comprise instructions to calculate the color score for a remainder of the plurality of database images that include a color, and

the instructions to determine the homography further comprise instructions to refine the homography based on the region of the color of the test image, if the test image is determined to include the color.

10. The device of claim 3, wherein,

the instructions to select further comprise instructions to combine the similarity scores of each of the similarity sets, to determine a total similarity score for each of plurality of database images, each of the total similarity scores to be normalized, and

the instructions to select further comprise instructions to compare a highest of the total similarity scores to a threshold, and

the instructions to select further comprise instructions to determine a type of the test image based on the database image corresponding to the highest total similarity score, if the highest total similarity score is greater than the threshold.

11. The device of claim 10, wherein,

the instructions to select further comprise instructions to look up each of the similarity scores at a scoring table to determine a corresponding probability score,

the probability score of the similarity score is to infer that test image is a same type as the corresponding database image, if the probability score is positive,

the probability score of the similarity score is to infer that test image is not a same type as the corresponding database image, if the probability score is negative, and

the instructions to select further comprise instructions to total the probability scores corresponding to each of the similarity sets to determine the normalized total similarity scores.

12. A method, comprising:

determining a homography between a test image and a first database image of a plurality of database images based on main line of text of the test image;

calculating a sub-read score on the test image based on the homography and a location of a sub-read at the first database image, if the first database image includes a sub-read;

calculating a logo score on the test image based on the homography and a location of a logo at the first database image, if the first database image includes a logo;

calculating a color score on the test image based on the homography and a location of a color at the first database image, if the first database image includes a color;

combining at least one of the sub-read, logo and color scores to determine a total similarity score for the first database image;

repeating the determining, calculating the sub-read score, calculating the logo score, calculating the color score and the combining for a remainder of the plurality of database images; and

selecting a type of the test image based on a highest total similarity score of the plurality of total similarity scores of the plurality of database images.

13. The method of claim 12, wherein,

the test image includes a vehicle registration plate,

the plurality of database images include different types of vehicle registration plates, and

at least one of the sub-read, logo and color scores relate to characteristics of the vehicle registration plate.

14. A non-transitory computer-readable storage medium storing instructions that, if executed by a processor of a device, cause the processor to:

determine a homography between a test image and a plurality of different types of database images based on a main line of text of the test image;

calculate a total similarity score for each of the different types of database images based on at least one of a sub-read, logo and color score for each of the different types of database images; and

select a type of the test image based on a highest total similarity score of the plurality of different types of database images, wherein

at least one of the sub-read, logo and color scores are determined based on the homography.

15. The non-transitory computer-readable storage medium of claim 14, wherein,

the main line of text of the test image includes a plurality of characters, and

the homography is determined based on a series of point correspondences between a top and a bottom of at least two of the plurality of characters of the main line of text in the test image and a normalized coordinate system.