SELECTIVE DIFFUSION OF FILTERED EDGES IN IMAGES

• US 20110229029A1
• Filed: 01/19/2011
• Published: 09/22/2011
• Est. Priority Date: 01/19/2010
• Status: Active Grant

• Current Assignee: Pixar (The Walt Disney Company)
• Original Assignee: Pixar (The Walt Disney Company)
• Inventors: Kass, Michael
• Granted Patent: US 8,478,064 B2
• Time in Patent Office: Days
• US Class Current: 382/168
• CPC Class Codes: G06T 2207/20012 Locally adaptive
• CPC Class Codes: G06T 2207/20028 Bilateral filtering
• CPC Class Codes: G06T 5/002 Denoising; Smoothing noise processing or correction adapted to be used in an image pickup device containing and electronic image sensor H04N23/81, H04N25/60 - H04N25/67
• CPC Class Codes: G06T 5/20 using local operators
• CPC Class Codes: G06T 5/70 Denoising; Smoothing

Abstract

An edge-preserving diffusion filter maintains the sharp edges in images while smoothing out image noise. An edge-preserving diffusion filter applies an edge-preserving smoothing filter to an image to form a filtered image. The modified image is then blurred by a blurring filter to form a blurred image. The modified image and the blurred image are blended together to form an output image based on an error metric associated with each pixel. The edge-preserving diffusion filter may be utilized to perform a multilevel decomposition of the image. The edge-preserving diffusion filter may be applied to an unfiltered image to produce a base image. The difference between the unfiltered image and the base image defines a detail image. The detail image may be used as the input for recursively generating additional levels of detail. The multilevel decomposition may utilize filter kernels associated with different contrast levels for each iteration.

Claims

1. A method of analyzing an image, the method comprising:

receiving at least one unfiltered image and at least one corresponding filtered image;

selecting a first image element of the filtered image;

selecting a first blurring operator;

applying the first blurring operator to a neighborhood surrounding the first image element to determined a blurred image element, wherein the blurring operator is adapted to reduce a magnitude of a high spatial frequency component of an image spectrum relative to a magnitude of a low spatial frequency component of the image spectrum;

evaluating an error metric to compare the blurred image element and the first image element with a corresponding image element of the unfiltered image;

generating an output image including at least a first output image element; and

in response to the error metric satisfying a first criteria, setting the first output image element to equal the blurred image element.

2. The method of claim 1, wherein the error metric is adapted to quantify deviations of image elements from corresponding image elements in the unfiltered image.

3. The method of claim 1, wherein:

in response to the error metric satisfying a second criteria, setting the output image element to equal the first image element.

4. The method of claim 1, wherein:

in response to the error metric satisfying a second criteria, blending the first image element with the blurred image element to form a blended image element and setting the first output image element equal to the blended image element.

5. The method of claim 1, wherein the blurring operator is a linear filter kernel.

6. The method of claim 1, wherein the first blurring operator is selected from a plurality of blurring operators.

7. The method of claim 6, wherein the plurality of blurring operators including a Gaussian blurring operator.

8. The method of claim 1, further comprising:

selecting a second blurring operator;

applying the second blurring operator to a second neighborhood surrounding the first output image element in the output image to determine a second blurred image element;

evaluating a second error metric to compare the second blurred image element and the first image element with the first output image element in the output image;

generating a second output image including at least a second output image element; and

in response to the error metric satisfying the first criteria, setting the second output image element to equal the second blurred image element.

9. The method of claim 1, wherein the first and second blurring operators have different neighborhood sizes.

10. The method of claim 1, wherein the filtered image is the result of the application of a local histogram-based filter.

11. The method of claim 1, wherein the filtered image is the result of the application of a bilateral filter.

12. The method of claim 1, wherein the error metric is based on a ratio of differences between each of the blurred image element and the first image element with respect to the corresponding image element.

13. The method of claim 1, wherein the unfiltered image is included in a video sequence.

14. A method of analyzing an image, the method comprising:

receiving at least one unfiltered image;

applying a first edge-preserving diffusion filter to the unfiltered image to form a first base image;

generating a first detail image based on differences between the unfiltered image and the first base image; and

outputting the first base image and at least one detail image.

15. The method of claim 14, the method comprising:

applying a second edge-preserving diffusion filter to the first detail image to form a second base image;

generating a second detail image based on differences between the unfiltered image and the second base image; and

outputting the second base image.

16. The method of claim 15, wherein the output of at least one detail image includes the second detail image.

17. The method of claim 15, wherein the first edge-preserving diffusion filter is applied with a first filter kernel associated with a first contrast level and the second edge-preserving diffusion filter is applied with a second filter kernel associated with a second contrast level.

18. The method of claim 17, wherein the first and second filter kernels are associated with a first neighborhood size.

19. The method of claim 17, wherein the at least one of the first and second filter kernels is a Gaussian filter kernel.

20. The method of claim 15, wherein at least one of the first and second edge-preserving diffusion filters includes an application of a local histogram-based filter.

21. The method of claim 15, wherein at least one of the first and second edge-preserving diffusion filters includes an application of a bilateral filter.

22. A method of analyzing an image, the method comprising:

receiving a first image;

applying a first filter to the first image to create a filtered image;

applying a first blurring operator to the filtered image to create a blurred image;

comparing each of the filtered image and the blurred image with the first image to identify at least a first portion of the blurred image that is closer to the first image than the filtered image and at least a second portion of the filtered image that is closer to the first image than the blurred image; and

creating an output image including the first portion of the blurred image and the second portion of the filtered image.

23. The method of claim 22, wherein the blurring operator is adapted to reduce a magnitude of a high spatial frequency component of an image spectrum relative to a magnitude of a low spatial frequency component of the image spectrum.

24. The method of claim 22, wherein comparing each of the filtered image and the blurred image with the first image comprises:

determining a first error metric value for a first image element included in the blurred image based on a difference between the first image element and a corresponding second image element included in the first image;

determining a second error metric value for a third image element included in the filtered image based on a difference between the third image element and the second image element included in the first image, wherein the first image element in the blurred image corresponds with the second image element included in the filtered image;

determining if the first image element or the third image element is closer to the second image element in the first image based on the first and second error metric values;

including the first image element in the first portion of the blurred image in response to the first image element being closer to the second image element in the first image; and

including the third image element in the second portion of the filtered image in response to the third image element being closer to the second image element in the first image.

25. The method of claim 22, comprising:

identifying third portion of the blurred image and a fourth portion of the filtered image having similar deviations from the first image;

blending the third portion of the blurred image with the fourth portion of the filtered image to form a blended image portion; and

including the blended image portion in the output image.

26. The method of claim 22, wherein the first blurring operator includes a Gaussian blurring operator.

27. The method of claim 22, further comprising:

applying a second blurring operator to the output image to create a second blurred image;

comparing each of the output image and the second blurred image with the first image to identify at least a third portion of the second blurred image that is closer to the first image than the output image and a fourth portion of the output image that is closer to the first image than the second blurred image;

creating a second output image including the third portion of the second blurred image and the fourth portion of the output image.

28. The method of claim 27, wherein the first blurring operator is associated with a first neighborhood having a first size and the second blurring operator is associated with a second neighborhood having a second size different than the first size.

29. The method of claim 22, wherein the first filter includes a local histogram-based filter.

30. The method of claim 22, wherein the first filter includes a bilateral filter.