Bayesian methods for noise reduction in image processing

• US 7,813,581 B1
• Filed: 05/06/2005
• Issued: 10/12/2010
• Est. Priority Date: 05/06/2005
• Status: Expired due to Fees

• Current Assignee: Allen Robert Tannenbaum, Ben G. Fitzpatrick
• Original Assignee: Allen Robert Tannenbaum, Ben G. Fitzpatrick
• Inventors: Tannenbaum, Allen Robert, Fitzpatrick, Ben G.
• Primary Examiner(s): Bhatnagar; Anand
• Assistant Examiner(s): Torres; Jose M
• Application Number:
  US11/123,445
• Time in Patent Office: 1,985 Days
• Field of Search: 382/276
• US Class Current: 382/260
• CPC Class Codes: G06T 5/70 Denoising; Smoothing
• CPC Class Codes: H04N 25/60 Noise processing, e.g. detecting, correcting, reducing or removing noise

Abstract

Improved methodology for image processing and object tracking that, inter alia, reduces noise. In one embodiment, the methodology is applied to moving targets, and comprises processing sequences of images that have been corrupted by one or more noise sources (e.g., sensor noise, medium noise, and/or target reflection noise). A likelihood or similar logical construct (e.g., Bayes'"'"' rule) is applied to the individual images (or aggregations thereof) of an image sequence in order to generate a posterior image for each observed image. The posterior images are fed-forward to the determination of the posterior image for one or more subsequent images (after smoothing), thereby making these subsequent determinations more accurate. The net result is a more accurate and noise-reduced representation (and location) of the target in each image.

Claims

1. A method of reducing noise associated with a series of images obtained from a moving target, comprising:

providing at least first and second frames of image data, each of said frames comprising a plurality of pixels;

for said first frame, calculating the posterior probability that each of said plurality of pixels is a pixel of said target based at least in part on a prior probability;

forming a posterior probability image based at least in part on said act of calculating;

forming a reduced noise image by applying a filtering operation to said posterior probability image for use as a prior probability for at least said second frame; and

providing said reduced noise image to a tracking system.

2. The method of claim 1, wherein said noise comprises at least one of sensor-related noise, medium-related noise and target reflection noise.

3. The method of claim 1, wherein said act of providing said at least first and second frames of image data comprises:

illuminating said target with an energy source; and

obtaining said at least first and second frames using a sensor adapted to detect energy emitted by said source and reflected off of said target.

4. The method of claim 1, wherein said act of calculating comprises applying a Bayesian function to at least a portion of said prior probability.

5. The method of claim 4, wherein said calculating is based at least in part on the intensity level associated with a given one of said pixels.

6. The method of claim 1, wherein said filtering at least a portion of said posterior probability image comprises performing a Gaussian-smoothing convolution.

7. The method of claim 1, wherein said filtering at least a portion of said posterior probability image comprises performing a nonlinear propagation.

8. A method of reducing noise in images, comprising:

for a plurality of pixels of a first image, calculating a posterior probability (π

⁺) that each pixel (i, j) comprises a target pixel given an image observed at a first time using a prior probability value;

forming a posterior probability image based on said posterior probabilities for each of said pixels;

smoothing said posterior probability image by filtering at least portions of said posterior image to produce a reduced noise image, and storing said reduced noise image for use as a prior probability for an image observed at a time subsequent to said first time; and

providing said reduced noise image to a tracking system.

9. A method of reducing noise associated with a series of images obtained from a moving target, comprising:

providing at least first and second frames of image data, each of said frames comprising a plurality of pixels;

for said first frame, calculating the posterior probability that each of said plurality of pixels is a pixel of said target by applying a Bayesian function based at least in part on a prior probability;

forming a posterior probability image based at least in part on said act of calculating;

forming a reduced noise images by applying a filtering operation comprising a Gaussian-smoothing convolution to said posterior probability image for use as a prior probability for at least said second frame; and

providing said reduced noise images to a tracking system.

10. The method of claim 9, wherein said noise comprises at least one of sensor-related noise, medium-related noise and/or target reflection noise.

11. The method of claim 9, wherein said act of providing said at least first and second frames of image data comprises:

illuminating said target with an energy source; and

obtaining said at least first and second frames using a sensor adapted to detect energy emitted by said source and reflected off of said target.

12. The method of claim 9, wherein said calculating is based at least in part on the intensity level associated with a given one of said pixels.

13. A method of sequentially processing frames of image data comprising a plurality of pixels, said method comprising:

assigning likelihoods of said pixels belonging to one of a plurality of classes;

processing said image data based at least in part on said act of assigning likelihoods; and

utilizing said processed image data in a tracking system;

wherein said act of assigning likelihoods comprises assigning a likelihood to at least a first frame of image data, and said act of processing further comprises processing at least portions of said first frame of image data for use in processing at least one subsequent frame of image data; and

wherein said use in processing said at least one subsequent frame of image data comprises performing a mapping of posterior data to future prior data using at least one of a Gaussian-smoothing convolution or nonlinear propagation.

14. A method of reducing noise associated with a series of images obtained from a moving target, comprising:

providing at least first and second frames of image data, each of said frames comprising a plurality of pixels;

for said first frame, calculating the posterior probability that each of said plurality of pixels is a pixel of said target based at least in part on a prior probability;

forming a posterior probability image based at least in part on said posterior probability; and

forming a reduced noise image by processing said posterior probability image for use as a prior probability for at least said second frame;

wherein said reduced noise image is provided to a system capable of using said reduced noise image in providing a function relating to said target.

15. A method of reducing noise associated with a series of images obtained from a moving target, comprising:

a step for providing at least first and second frames of image data, each of said frames comprising a plurality of pixels;

a step for, for said first frame, calculating the posterior probability that each of said plurality of pixels is a pixel of said target based at least in part on a prior probability;

a step for forming a posterior probability image based at least in part on said posterior probability;

a step for forming a reduced noise image by processing said posterior probability image for use as a prior probability for at least said second frame; and

a step for providing said reduced noise image to a system capable of using said reduced noise image in providing a function relating to said target.