Fractal-based image compression and interpolation

• US 5,148,497 A
• Filed: 10/10/1991
• Issued: 09/15/1992
• Est. Priority Date: 02/14/1990
• Status: Expired due to Term

• Current Assignee: The Analytic Science Corporation
• Original Assignee: Massachusetts Institute of Technology
• Inventors: Stephenson, Thomas P., Simoncelli, Eero P., Pentland, Alex P.
• Primary Examiner(s): Boudreau, Leo H.
• Assistant Examiner(s): Johns, Andrew W.
• Application Number:
  US07/773,779
• Time in Patent Office: 341 Days
• Field of Search: 382/41, 382/47, 382/54, 382/56, 364/723, 364/724.1, 395/128
• US Class Current: 382/249
• CPC Class Codes: G06F 17/175 of multidimensional data
• CPC Class Codes: H04N 19/63 using sub-band based transform, e.g. wavelets
• CPC Class Codes: H04N 19/99 involving fractal coding

Abstract

An image interpolator for converting an original image into an enhanced image, the interpolator including rule logic for embodying a rule specifying how pixel patterns associated with first subband image data transform into corresponding pixel patterns associated with second subband image data, the first and second subband image data being derived from a reference image; and conversion logic for generating the enhanced image from the original image based upon the rule.

Claims

1. An image interpolator for converting an original image into an enhanced image, the interpolator comprising:

rule logic for embodying a rule specifying how pixel patterns associated with first subband immage data transform into corresponding pixel patterns associated with second subband image data, said first and second subband image data being spatial frequency subband image data derived from a reference image that is generated by a single sensor; and

conversion logic for generating the enhanced image from the original image based upon said rule, wherein said conversion logic comprises;

means for generating high resolution subband image data from the second subband image data in accordance with said rule; and

means for combining the original image with the high resolution image data to generate the enhanced image.

2. The image interpolator of claim 1 wherein said reference image is derived from the original image.

3. The image interpolator of claim 2 wherein said reference image is the original image or a subset of the original image.

4. The image interpolator of claim 3 wherein the original image is decomposable into a pyramid of bandpass images and said first and second subband image data correspond to different images of said pyramid.

5. The image interpolator of claim 4 wherein said first and second subband image data correspond to neighboring images of said pyramid.

6. The image interpolator of claim 5 wherein successive images of said pyramid of bandpass images are scaled versions of the immediately preceding image from which it was derived.

7. The image interpolator of claim 6 wherein a scaling factor of R relates neighboring levels of the image pyramid, R being a positive integer.

8. The image interpolator of claim 7 wherein R is equal to 2.

9. The image interpolator of claim 2 wherein said first and second subband image data correspond to neighboring subbands of the reference image.

10. The image interpolator of claim 1 wherein said second subband image data represents higher spatial frequencies than said first subband image data.

11. The image interpolator of claim 1 wherein said high resolution subband image data has a spatial frequency resolution that is higher than the spatial frequency resolution of the original image.

12. The image interpolator of claim 1 wherein said rule logic further comprises:

a filter for generating said first and second subband image data from the original image; and

means for deriving said rule from said first and second subband image data.

13. The image interpolator of claim 12 wherein said first and second subband image data are derived from the original image, said first and second subband image data corresponding respectively to a first and a second subband image;

and said rule deriving means comprises table generating logic for generating a table relating each of a plurality of N×

N pixel patterns from said first subband image to a corresponding M×

M pixel pattern of said second subband image.

14. The image interpolator of claim 13 wherein said conversion logic comprises:

means for generating high resolution subband image data from the second subband image data in accordance with said rule, said high resolution subband image data having a spatial frequency resolution that is higher than the spatial frequency resolution of the original image; and

means for combining the original image with the high resolution image data to generate the enhanced image.

15. The image interpolator of claim 13 wherein said conversion logic comprises:

selection logic for selecting an N×

N pixel pattern of the second subband image;

search logic for finding an N×

N pixel pattern in the table which best matches the selected N×

N pixel pattern; and

assignment logic for assigning to corresponding locations within the high resolution subband image data an M×

M pixel pattern from the table that is associated with the identified N×

N pixel pattern.

16. The image interpolator of claim 12 wherein said first and second subband image data are derived from the original image, said first and second subband image data corresponding respectively to a first and a second subband image;

and said rule deriving means comprises neural network generating logic for generating an associative neural network relating each of a plurality of N×

N pixel patterns from said first subband image to a corresponding M×

M pixel pattern of said second subband image.

17. The image interpolator of claim 12 wherein said first and second subband image data are derived from the original image, said first and second subband image data corresponding respectively to a first and a second subband image;

and said rule deriving means comprises coefficient generating logic for generating coefficients for a set of equations relating each of a plurality of N×

N pixel patterns from said first subband image to a corresponding M×

M pixel pattern of said second subband image.

18. The image interpolator of claim 12 wherein said filter employs a Gaussian-like filter to generate said first and second subband image data.

19. The image interpolator of claim 12 wherein said filter employs a quadrature mirror filter to generate said first and second subband image data.

20. A method for converting an original image into an enhanced image, the method comprising:

determining a rule according to which pixel patterns associated with first subband image data are transformed into corresponding pixel patterns associated with second subband image data, said first and second subband image data being spatial frequency subband image data derived from a reference image that is generated by a single sensor; and

generating the enhanced image from the original image based upon said rule, wherein said step of generating the enhanced image comprises;

generating high resolution subband image data from the second subband image data in accordance with said rule; and

combining the original image with the high resolution image data to generate the enhanced image.

21. The method of claim 20 wherein said reference image is derived from the original image.

22. The method of claim 21 wherein said reference image is the original image or a subset of the original image.

23. The image interpolator of claim 22 wherein the original image is decomposable into a pyramid of bandpass images and said first and second subband image data correspond to different images of said pyramid.

24. The image interpolator of claim 23 wherein said first and second subband image data correspond to neighboring images of said pyramid.

25. The image interpolator of claim 24 wherein successive images of said pyramid of bandpass images are scaled versions of the immediately preceding image from which it was derived.

26. The image interpolator of claim 25 wherein a scaling factor or R relates neighboring levels of the image pyramid, R being a positive integer.

27. The image interpolator of claim 26 wherein R is equal to 2.

28. The image interpolator of claim 21 wherein said first and second subband image data correspond to neighboring subbands of the reference image.

29. The method of claim 20 wherein said second subband image data represents higher spatial frequencies than said first subband image data.

30. The method of claim 20 wherein said high resolution subband image data has a spatial frequency resolution that is higher than the spatial frequency resolution of the original image.

31. The method of claim 20 wherein said rule determining step further comprises:

filtering said original image to generate said first and second subband image data; and

deriving said rule from said first and second subband image data.

32. The method of claim 31 wherein said first and second subband image data correspond respectively to a first and a second subband image;

and said rule deriving step comprises generating a table relating each of a plurality of N×

N pixel patterns from said first subband image to a corresponding M×

M pixel pattern of said second subband image.

33. The method of claim 32 wherein said enhanced image generating step comprises:

generating high resolution subband image data from the second subband image data in accordance with said rule, said high resolution subband image data having a spatial frequency resolution that is higher than the spatial frequency resolution of the original image; and

combining the original image with the high resolution image data to generate the enhanced image.

34. The method of claim 32 wherein said enhanced image generating step comprises:

selecting an N×

N pixel pattern of the second subband image;

finding an N×

N pixel pattern in the table which best matches the selected N×

N pixel pattern; and

assigning to corresponding locations within the high resolution subband image data an M×

M pixel pattern from the table that is associated with the identified N×

N pixel pattern.

35. The method of claim 31 wherein said filtering step employs a Gaussian-like filter to generate said first and second subband image data.

36. The method of claim 31 wherein said filtering step employs a quadrature mirror filter to generate said first and second subband image data.