Character recognition in distorted images

• US 8,989,446 B2
• Filed: 01/18/2012
• Issued: 03/24/2015
• Est. Priority Date: 01/18/2011
• Status: Expired due to Fees

• Current Assignee: RTC Vision, Ltd.
• Original Assignee: RTC Vision, Ltd.
• Inventors: Kagarlitsky, Sefy, Rubner, Joseph, Avrahami, Nir, Falik, Yohai
• Primary Examiner(s): Lefkowitz, Sumati
• Assistant Examiner(s): JAMES, THOMAS AIDEN
• Application Number:
  US13/993,276
• Publication Number:
  US 20130279759A1
• Time in Patent Office: 1,161 Days
• Field of Search: 382/105
• 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 comprising:

(a) prior to rendering an imitation image, estimating one or more rendering parameters from a distorted image;

(b) rendering, by a processor the imitation image, the imitation image comprising one or more simulated characters that correspond to one or more characters within the distorted image;

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

(d) comparing the distorted imitation image with the distorted image to compute a similarity between the distorted imitation image and the distorted image; and

(e) identifying the one or more characters as the one or more simulated characters based on the similarity.

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

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

4. The method of claim 1, 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, or metadata relating to at least one of the characters and the distorted image.

5. The method of claim 1, further comprising:

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

6. The method of claim 5, wherein the metadata comprises at least one of a compression type and compression parameters.

7. The method of claim 1, further comprising:

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

8. The method of claim 7, 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.

9. The method of claim 1, further comprising:

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

11. The method of claim 1, wherein the one or more simulated characters account for at least one of:

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

12. 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, or a quantization model.

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

repeating (b)-(d) for each of a plurality of distorted images.

14. The method of claim 13, further comprising:

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

15. The method of claim 13, further comprising:

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

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

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

17. The method of claim 1, wherein the similarity comprises a probability that the one or more simulated characters are present in the distorted image.

18. The method of claim 13, wherein the similarity comprises a probability that the one or more simulated characters are present in the distorted images.

19. A system comprising one or more processors configured to:

(a) render an imitation image, the imitation image comprising one or more simulated characters that correspond to one or more characters within a distorted image;

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

(c) compare 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, the similarity comprising a probability that the one or more simulated characters are present in the distorted image; and

(d) identify the one or more characters as the one or more simulated characters based on the similarity.

20. A non-transitory computer-readable medium comprising instructions that, when executed by a processing device, cause the processing device to:

(a) receive a distorted data element;

(b) render a plurality of imitation data elements, each of the imitation data elements comprising a unique undistorted simulation of the distorted data element;

(c) determine 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, or

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

(e) compare 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) compare each of the computed similarities with one another to identify a most accurate simulation; and

(g) output at least one of;

each of the computed similarities or the most accurate simulation.

21. The non-transitory computer-readable medium of claim 20, wherein the distorted data element comprises at least one of an audio file or a three-dimensional model.