SYSTEM AND METHOD FOR IMPROVED CHARACTER RECOGNITION IN DISTORTED IMAGES

• US 20130279759A1
• Filed: 01/18/2012
• Published: 10/24/2013
• Est. Priority Date: 01/18/2011
• Status: Active Grant

• Current Assignee: RTC Vision, Ltd.
• Original Assignee: RTC Vision, Ltd.
• Inventors: Rubner, Joseph, Avrahami, Nir, Falik, Yohai, Kagarlitsky, Sefy
• Granted Patent: US 8,989,446 B2
• Time in Patent Office: Days
• US Class Current: 382/105
• CPC Class Codes: G06V 30/10 Character recognition
• CPC Class Codes: G06V 30/1902 Shifting or otherwise transforming the patterns to accommodate for positional errors

Abstract

Systems and methods are provided for recognizing characters within a distorted image. According to a one aspect, a method for recognizing one or more characters within a distorted image includes rendering one or more imitation images, the imitation images including simulations of the distorted image, applying one or more distortion models to the imitation images, thereby generating distorted imitation images, comparing the distorted imitation images with the distorted image in order to compute similarities between the distorted imitation images and the distorted image, and identifying the characters based on the best similarity. According to other aspects, the systems and methods can be configured to provide recognition of other distorted data types and elements.

Claims

1. A method for recognizing one or more characters within a distorted image using a computing device, the method comprising:

(a) rendering, with a processor executing code, an imitation image;

(b) applying one or more distortion models to the imitation image, thereby generating a distorted imitation image;

(c) comparing the distorted imitation image with the distorted image in order to compute a similarity between the distorted imitation image and the distorted image; and

(d) identifying the characters based on the similarity.

2. The method of claim 1, wherein the imitation image is rendered based on at least one defined standard.

3. The method of claim 2, wherein the defined standard is a license plate standard.

4. The method of claim 1, further comprising:

prior the rendering step, estimating one or more rendering parameters from the distorted image.

5. The method of claim 4, wherein the rendering parameters are estimated based on at least one of:

an estimation of the characters, prior data relating to at least one of the characters and the distorted image, and metadata relating to at least one of the characters and the distorted image.

6. The method of claim 4, wherein the imitation image comprises one or more simulated characters that correspond to the one or more characters within the distorted image, the identifying step further comprising identifying the characters as the simulated characters based on the similarity.

7. The method of claim 1, further comprising:

prior the applying step, determining the one or more distortion models to be applied to the imitation image based on metadata that corresponds to the distorted image.

8. The method of claim 7, wherein the metadata comprises at least one of a compression type and compression parameters.

9. The method of claim 1, further comprising:

processing the distorted image in order to determine one or more distortion types present in the distorted image.

10. The method of claim 9, wherein the one or more distortion models applied to the imitation image correspond to the one or more distortion types present in the distorted image.

11. The method of claim 1, further comprising:

processing one or more images related to the distorted image in order to determine one or more distortion types present in the one or more images related to the distorted image.

12. The method of claim 11, wherein the one or more distortion models applied to the imitation image correspond to the one or more distortion types present in the one or more images related to the distorted image.

13. The method of claim 1, wherein the imitation image comprises one or more simulated characters that correspond to the one or more characters within the distorted image, the one or more simulated characters accounting for at least one of:

viewing angle, color, size, space between characters and imaging model.

14. The method of claim 1, wherein the one or more distortion models comprise at least one of:

a blur model, a lossy compression model, an interlacing model, a pixelization model, a image noise model, an atmospheric turbulence model, and a quantization model.

15. The method of claim 1, wherein the distorted image comprises a plurality of distorted images, the method further comprising:

repeating steps (a)-(c) for each of a plurality of distorted images.

16. The method of claim 15, further comprising:

comparing each of the computed similarities with one another in order to identify a most accurate simulation.

17. The method of claim 15, further comprising:

combining the computed similarities in order to identify a most accurate simulation.

19. The method of claim 1, wherein the distorted image comprises a plurality of distorted images, the method further comprising:

repeating steps (a)-(b) for each of a plurality of distorted images, wherein the comparing step comprises comparing each of the distorted imitation images generated at the each respective iteration of step (b) with the distorted images in order to compute a composite similarity across the distorted imitation images and the distorted images.

20. The method of claim 1, wherein the imitation image comprises one or more simulated characters that correspond to the one or more characters within the distorted image, and wherein the similarity comprises a probability that the one or more simulated characters are present in the distorted image.

21. The method of claim 15, wherein the imitation image comprises one or more simulated characters that correspond to the one or more characters within the distorted image, and wherein the similarity comprises a probability that the one or more simulated characters are present in the distorted images.

18. (canceled)

22-24. -24. (canceled)

25. A method for recognizing one or more characters within a distorted image using a computing device, the method comprising:

(a) rendering, with a processor executing code, an imitation image;

(b) applying at least a first distortion model to the imitation image, thereby generating an initially distorted imitation image;

(c) comparing the initially distorted imitation image with the distorted image by

i. simulating a further distortion of the initially distorted imitation image, the further distortion comprising at least a second distortion model, and

ii. computing a similarity between the further distortion of the initially distorted imitation image and the distorted image; and

(d) identifying the characters based on the similarity.

26. A method for processing a distorted data element using a computing device, the method comprising:

(a) receiving the distorted data element;

(b) rendering, with a processor executing code, a plurality of imitation data elements, each of the imitation data elements comprising a unique undistorted simulation of the distorted data element;

(c) determining one or more distortion models to be applied to each of the imitation data elements, the one or more distortion models being applied based on at least one of;

i. metadata that corresponds to the distorted data element,

ii. a processing of the distorted data element in order to determine one or more distortion types present in the distorted data element, and

iii. a processing of one or more data elements related to the distorted data element in order to determine one or more distortion types present in the one or more data elements related to the distorted data element;

(d) applying the one or more distortion models to each of the imitation data elements, thereby generating a plurality of distorted imitation data elements;

(e) comparing each of the distorted imitation data elements with the distorted data element in order to compute a similarity between each respective distorted imitation data element and the distorted data element;

(f) comparing each of the computed similarities with one another in order to identify a most accurate simulation; and

(g) outputting at least one of;

each of the computed similarities and the most accurate simulation.

27. The method of claim 26, wherein the distorted data element comprises at least one of an audio file and a three-dimensional model.

28-33. -33. (canceled)