Image processing method for object recognition and dynamic scene understanding

• US 7,664,339 B2
• Filed: 05/03/2005
• Issued: 02/16/2010
• Est. Priority Date: 05/03/2004
• Status: Expired due to Fees

• Current Assignee: Jacek Turski
• Original Assignee: Jacek Turski
• Inventors: Turski, Jacek
• Primary Examiner(s): BHATNAGAR, ANAND P
• Application Number:
  US11/120,607
• Publication Number:
  US 20050244059A1
• Time in Patent Office: 1,750 Days
• Field of Search: 382/100, 382/276, 382/279, 382/280, 382/281, 382/210, 708/403, 708/404, 708/405, 708/821, 702/77, 370/210, 345/427, 345/604, 345/645, 345/70
• US Class Current: 382/280
• CPC Class Codes: G06F 17/141 Discrete Fourier transforms
• CPC Class Codes: G06T 3/00 Geometric image transformations in the plane of the image
• CPC Class Codes: G06T 7/262 using transform domain methods, e.g. Fourier domain methods
• CPC Class Codes: G06T 7/37 using transform domain methods
• CPC Class Codes: G06V 10/10 Image acquisition document image scanning and transmission H04N1/00; control of digital cameras H04N23/60
• CPC Class Codes: G06V 10/247 by affine transforms, e.g. correction due to perspective effects; Quadrilaterals, e.g. trapezoids
• CPC Class Codes: G06V 20/653 by matching three-dimensional models, e.g. conformal mapping of Riemann surfaces
• CPC Class Codes: G06V 2201/11 Technique with transformation invariance effect
• CPC Class Codes: G06V 2201/12 Acquisition of 3D measurements of objects

Abstract

Provided is a method for digital image representation based upon Discrete Projective Fourier Transform (DPFT) constructed in the noncompact (DNPFT) and compact (DCPFT) realizations of geometric Fourier analysis on SL(2,C) groups. Novel characteristics are that the model is well adapted to perspective image transformations and well adapted to the retinotopic (conformal) mapping of the biological visual system. To compute the DPFT of a digital image by Fast Fourier transform (FFT), an image is re-sampled with a non-uniform log-polar sampling geometry. A “deconformalization” procedure corrects the “conformal lens optics” of the conformal camera to render image perspective transformations. DNPFT computes the convolution in the noncompact realization defined over 2-dimensional rotations in the image plane and dilations while the DCPFT computes the convolution in the compact realization (which is defines over all 3-dimensional rotations) and therefore provides basis for developing projectively invariant under all rotations object matching.

Claims

1. A method of processing an image for pattern recognition and three-dimensional scene resolution, comprising:

receiving a first digitized image;

creating a first discrete projective Fourier transform (DPFT) image in log-polar coordinates of the first digitized image using a fast Fourier transform (FFT);

storing the first DPFT image in a memory;

projectively transforming the log-polar coordinates of the DPFT image; and

computing an inverse DPFT image of the projectively transformed log-polar coordinates of the DPFT image using a non-uniform FFT to produce a projective transformation of the first digitized image.

2. The method of claim 1, the receiving a first digitized image comprising:

capturing an analog image on an image plane; and

digitizing the captured image.

3. The method of claim 1, wherein the digitized image is produced by a silicon retina.

4. The method of claim 1, further comprising:

correcting the projective transformation for conformal distortions to obtain a an image perspective transformation.

5. The method of claim 1, further comprising:

receiving a second digitized image;

creating a second DPFT image in log-polar coordinates of the second digitized image using a FFT;

producing a product by multiplying the first DPFT image and the second DPFT image;

computing the an inverse DPFT of the product using FFT to create a projective convolution; and

determining whether or not the first image is a projectively independent match of the second image based upon the projective convolution.

6. The method of claim 5, further comprising determining a physical location relative to a capture location of the first image based upon a determination of a projectively independent match between the first image and the second image.

7. A system processing an image for pattern recognition and three-dimensional scene resolution, comprising:

a processor;

a memory coupled to the processor;

logic for receiving a first digitized image;

logic for creating a first discrete projective Fourier transform (DPFT) in log-polar coordinates of the first digitized image using a fast Fourier transform (FFT);

logic for storing the first DPFT image in the memory

logic for projectively transforming the log-polar coordinates of the DPFT image; and

logic for computing an inverse DPFT image of the projectively transformed log-polar coordinates of the DPFT image using a non-uniform FFT to produce a projective transformation of the first digitized image.

8. The system of claim 7, the logic for receiving a first digitized image comprising:

logic for capturing an analog image on an image plane; and

logic for digitizing the captured image.

9. The system of claim 7, further comprising a silicon retina wherein the digitized image is produced by the silicon retina.

10. The system of claim 7, further comprising logic for correcting the image transformation for conformal distortions to obtain an image perspective transformation.

11. The system of claim 7, further comprising:

logic for receiving a second digitized image;

logic for creating a second DPFT image in log-polar coordinates of the second digitized image using a FFT;

logic for producing a product by multiplying the first DPFT image and the second DPFT image;

logic for computing the an inverse DPFT of the product using FFT to create a projective convolution; and

logic for determining whether or not the first image is a projectively independent match of the second image based upon the projective convolution.

12. The system of claim 11, further comprising logic for determining a physical location relative to a capture location of the first image based upon a determination of a projectively independent match between the first image and the second image.

13. A computer programming product for processing an image for pattern recognition and three-dimensional scene resolution, comprising:

a computer-readable memory;

logic, stored on the memory, for receiving a first digitized image;

logic, stored on the memory, for creating a first discrete projective Fourier transform (DPFT) in log-polar coordinates of the first digitized image using a fast Fourier transform (FFT);

logic, stored on the memory, for storing the first DPFT image in the memory

logic, stored on the memory, for projectively transforming the log-polar coordinates of the DPFT image;

logic, stored on the memory, for computing an inverse DPFT image of the projectively transformed log-polar coordinates of the DPFT image using a non-uniform FFT to produce a projective transformation of the first digitized image.

14. The computer programming product of claim 13, the logic for receiving a first digitized image comprising:

logic, stored on the memory, for capturing an analog image on an image plane; and

logic, stored on the memory, for digitizing the captured image.

15. The computer programming product of claim 13, further comprising logic, stored on the memory, for correcting the image transformation for conformal distortions to obtain a an image perspective transformation.

16. The computer programming product of claim 13, further comprising:

logic, stored on the memory, for receiving a second digitized image;

logic, stored on the memory, for creating a second DPFT image in log-polar coordinates of the second digitized image using a FFT;

logic, stored on the memory, for producing a product by multiplying the first DPFT image and the second DPFT image;

logic, stored on the memory, for computing the an inverse DPFT of the product using FFT to create a projective convolution; and

logic, stored on the memory, for determining whether or not the first image is a projectively independent match of the second image based upon the projective convolution.

17. The computer programming product of claim 16, further comprising logic, stored on the memory, for determining a physical location relative to a capture location of the first image based upon determination of a projectively independent match between the first image and the second image.