Method and apparatus for removing image noise

• US 8,611,680 B2
• Filed: 11/11/2008
• Issued: 12/17/2013
• Est. Priority Date: 11/30/2007
• Status: Active Grant

• Current Assignee: Trustees Of Columbia University In The City Of New York (Columbia University)
• Original Assignee: Kyoto University
• Inventors: Yamamoto, Yutaka, Nagahara, Masaaki, Kobayashi, Akira
• Primary Examiner(s): Chen, Wenpeng
• Application Number:
  US12/745,011
• Publication Number:
  US 20110002552A1
• Time in Patent Office: 1,862 Days
• Field of Search: None
• US Class Current: 382/233
• CPC Class Codes: H04N 19/117 Filters, e.g. for pre-processing or post-processing sub-band filter banks H04N19/635
• CPC Class Codes: H04N 19/14 Coding unit complexity, e.g. amount of activity or edge presence estimation H04N19/146 takes precedence
• CPC Class Codes: H04N 19/176 the region being a block, e.g. a macroblock
• CPC Class Codes: H04N 19/436 using parallelised computational arrangements
• CPC Class Codes: H04N 19/86 involving reduction of coding artifacts, e.g. of blockiness

Abstract

An FFT analysis is performed on the decoded image data of one frame, and a cutoff frequency based on the assumption that the original signal has first-order attenuation characteristics is obtained in both the horizontal and vertical directions. A sampled-data H^∞ filter (digital filter), previously designed to have different parameters corresponding to different cutoff frequencies is selected. By using this filter, a filtering process is performed on the image data decoded both in the horizontal and vertical directions. Since the variety of the analog frequency characteristics of the original image is taken into account for every frame of the image, the information that seems to be noise in the original image is less likely to be removed, and mosquito noise and block noise associated with a coding/decoding process can be efficiently removed.

Claims

1. A method for removing image noise, which is generated when image data compression-encoded in a given format is decoded to reproduce an image, comprising:

a) a deduction step for examining a digital frequency response of image data composing an image of a complete image or a frame of an image, and which have been compression-encoded and decoded, to deduce analog frequency characteristics of the image before compression encoding, wherein in the deduction step, a cutoff frequency is calculated from FFT data obtained by performing an FFT on an image for one frame of a moving image, based on an assumption that the analog frequency characteristics of the image before compression encoding are the analog frequency characteristics of a lowpass filter having attenuation characteristics of a given order;

b) a parameter setting step for changing a parameter of a sampled-data H^∞

filter in accordance with a deduction result by the deduction step, the parameter being a filter coefficient; and

c) a processing step for performing a filtering process on the image data after the image data is decoded and before the image data is converted by using the sampled-data H^∞

filter having a parameter calculated by solving a computational equation with H^∞

control under a predetermined condition or with a rigorous calculation which corresponds to the H^∞

control, the computational equation being obtained by approximately converting a conditional equation to a finite-dimensional discrete-time system, and the conditional equation being set to design the sampled-data H^∞

filter so as to reduce an error between an original analog image signal and an analog image signal obtained through an analog to digital conversion, a coding and decoding process, and a digital to analog conversion.

2. The method for removing an image noise according to claim 1, wherein:

the filtering process in the processing step is performed to each of image data aligned in a horizontal direction and image data aligned in a vertical direction of an image of a complete image or a frame of image.

3. The method for removing an image noise according to claim 1, wherein:

in the deduction step, an image of a complete image or a frame of image is sectionalized into small areas in each of which the analog frequency characteristics in the image before compression encoding can be considered to be similar and the analog frequency characteristics are deduced for each of the small areas; and

in the parameter setting step, the parameter of the digital filter is changed for each of the small areas.

4. The method for removing an image noise according to claim 3, wherein:

in the deduction step, image data of an image of a complete image or a frame of image aligned in a horizontal direction or in a vertical direction are sequentially scanned, and a boundary of the small areas is defined where a variation of a value of the data is equal to or more than a predetermined threshold.

5. The method for removing an image noise according to claim 3, wherein:

in the deduction step, a cutoff frequency is calculated on an assumption that the analog frequency characteristics in the image before compression encoding are a lowpass filter having attenuation characteristics of a given order.

6. The method for removing an image noise according to claim 1, wherein the given order is a first order.

7. The method for removing an image noise according to claim 1, wherein:

in the processing step, the sampled-data H^∞

filter is combined with an ε

-separating nonlinear filter.

8. The method for removing an image noise according to claim 1, wherein a processing object is a still image which is compression-encoded in JPEG format.

9. The method for removing an image noise according to claim 1, wherein a processing object is a moving image which is compression-encoded in MPEG format.

10. An apparatus for removing image noise, which is generated when image data compression-encoded in a given format is decoded to reproduce an image, comprising:

a) a deduction means for examining a digital frequency response of image data composing an image of a complete image or a frame of an image, and which have been compression-encoded and decoded, to deduce analog frequency characteristics of the image before compression encoding, wherein in the deduction step, a cutoff frequency is calculated from FFT data obtained by performing an FFT on an image for one frame of a moving image, based on an assumption that the analog frequency characteristics of the image before compression encoding are the analog frequency characteristics of a lowpass filter having attenuation characteristics of a given order;

b) a parameter setting means for changing a parameter of a sampled-data H^∞

filter in accordance with a deduction result by the deduction means, the parameter being a filter coefficient; and

c) a processing means for performing a filtering process on the image data after the image data is decoded and before the image data is converted by using the sampled-data H^∞

filter having a parameter calculated by solving a computational equation with H^∞

control under a predetermined condition or with a rigorous calculation which corresponds to the H^∞

control, the computational equation being obtained by approximately converting a conditional equation to a finite-dimensional discrete-time system, and the conditional equation being set to design the sampled-data H^∞

filter so as to reduce an error between an original analog image signal and an analog image signal obtained through an analog to digital conversion, a coding and decoding process, and a digital to analog conversion.

11. The apparatus for removing an image noise according to claim 10, wherein:

the processing means performs the filtering process to each of image data aligned in a horizontal direction and image data aligned in a vertical direction of an image of a complete image or a frame of image.

12. The apparatus for removing an image noise according to claim 10, wherein:

the deduction means sectionalizes an image of a complete image or a frame of image into small areas in each of which the analog frequency characteristics in the image before compression encoding can be considered to be similar and deduces the analog frequency characteristics for each of the small areas; and

the parameter setting means changes the parameter of the digital filter for each of the small areas.

13. The apparatus for removing an image noise according to claim 12, wherein:

the deduction means sequentially scans image data of an image of a complete image or a frame of image aligned in a horizontal direction or in a vertical direction, and a boundary of the small areas is defined where a variation of a value of the data is equal to or more than a predetermined threshold.

14. The apparatus for removing an image noise according to claim 12, wherein:

the deduction means calculates a cutoff frequency on an assumption that the analog frequency characteristics in the image before compression encoding are a lowpass filter having attenuation characteristics of a given order.

15. The apparatus for removing an image noise according to claim 10, wherein:

the deduction means calculates a cutoff frequency on an assumption that the analog frequency characteristics in the image before compression encoding are a lowpass filter having attenuation characteristics of a given order.

16. The apparatus for removing an image noise according to claim 15, wherein the given order is a first order.

17. The apparatus for removing an image noise according to claim 10, wherein:

the processing means combines the sampled-data H^∞

filter with an s-separating nonlinear filter.

18. The apparatus for removing an image noise according to claim 10, wherein a processing object is a still image which is compression-encoded in JPEG format.

19. The apparatus for removing an image noise according to claim 10, wherein a processing object is a moving image which is compression-encoded in MPEG format.