Methods and systems for estimation of additive noise
Aspects of the present invention are related to systems and methods for estimation of additive noise in an image or in a video sequence. An additive-noise estimate may be computed based on a first significant peak in a histogram of standard-deviation values of patches of an image-channel image associated with an input image.
1. A method for estimating additive noise in an image, said method comprising:
computing a statistical feature value for each image patch in a plurality of image patches associated with an image-channel image;
forming a histogram of said computed statistical feature values;
identifying a first significant peak in said histogram;
computing a peak-width measure of the identified first significant peak; and
determining an additive noise estimate based on a comparison of the computed peak-width measure and a width threshold.
2. The method as described in claim 1 , wherein said statistical feature is a statistical feature selected from the group consisting of standard deviation, variance and sum of absolute differences.
3. The method as described in claim 1 , wherein said each image patch in said plurality of image patches is an overlapping image patch.
4. The method as described in claim 1 , wherein said each image patch in said plurality of image patches meets a first criterion.
5. The method as described in claim 1 , further comprising computing said image-channel image.
6. The method as described in claim 1 , wherein said forming said histogram comprises smoothing an initial histogram.
7. The method as described in claim 1 , wherein said image-channel image is an image selected from the group consisting of a luminance image, an R-channel image, a G-channel image, a B-channel image and a chrominance image.
8. The method as described in claim 1 , wherein said computing a peak-width measure comprises:
determining a first bin location, in said histogram, located prior to said histogram-bin location of said first significant peak, whereat a histogram count falls below a predetermined threshold; and
subtracting said first bin location from said histogram-bin location of said first significant peak.
9. The method as described in claim 1 , wherein forming a histogram includes accumulating the statistical feature values for each image patch in the plurality of image patches.
10. The method as described in claim 1 , wherein the first significant peak is a significant peak whereat the bin location of the significant peak corresponds to a smallest statistical feature compared to other significant peaks.
11. The method as described in claim 10 , wherein a peak is considered significant when a number of histogram counts within a significance-determination window centered at a peak location meets a first criterion.
12. The method as described in claim 1 , wherein determining an additive noise estimate based on a comparison of the computed peak-width measure and a width threshold includes setting the additive noise estimate to a default value, if the computed peak-width measure is greater than the width threshold.
13. The method as described in claim 12 , wherein said default value is a noise value associated with a previous frame in an image sequence associated with said image-channel image.
14. The method as described in claim 1 , wherein determining an additive noise estimate based on a comparison of the computed peak-width measure and a width threshold includes computing an additive-noise estimate using a histogram-bin location of said first significant peak and said peak-width measure, if the computed peak-width measure is less than the width threshold.
15. A non-transitory computer-readable medium encoded with a computer program code for implementing a method for estimating additive noise in an image, said method comprising:
computing a statistical feature value for each image patch in a plurality of image patches associated with an image-channel image;
forming a histogram of said computed statistical feature values;
identifying a first significant peak in said histogram;
computing a peak-width measure of the identified first significant peak; and
determining an additive noise estimate based on a comparison of the computed peak-width measure and a width threshold.
16. The non-transitory computer-readable medium as described in claim 15 , wherein said each image patch in said plurality of image patches meets a first criterion.
17. The non-transitory computer-readable medium as described in claim 15 , wherein said each image patch in said plurality of image patches is an overlapping image patch.
18. The non-transitory computer-readable medium as described in claim 15 , wherein said method further comprises computing said image-channel image.
19. The non-transitory computer-readable medium as described in claim 15 , wherein said forming said histogram comprises smoothing an initial histogram.
20. The non-transitory computer-readable medium as described in claim 15 , wherein said computing a peak-width measure comprises:
determining a first bin location, in said histogram, located prior to said histogram-bin location of said first significant peak, whereat a histogram count falls below a predetermined threshold; and
subtracting said first bin location from said histogram-bin location of said first significant peak.
21. The non-transitory computer-readable medium as described in claim 15 , wherein said statistical feature is a statistical feature selected from the group consisting of standard deviation, variance and sum of absolute differences.
22. The non-transitory computer-readable medium as described in claim 15 , wherein said image-channel image is an image selected from the group consisting of a luminance image, an R-channel image, a G-channel image, a B-channel image and a chrominance image.
23. The non-transitory computer-readable medium as described in claim 15 , wherein forming a histogram includes accumulating the statistical feature values for each image patch in the plurality of image patches.
24. The non-transitory computer-readable medium as described in claim 15 , wherein the first significant peak is a significant peak whereat the bin location of the significant peak corresponds to a smallest statistical feature compared to other significant peaks.
25. The non-transitory computer-readable medium as described in claim 24 , wherein a peak is considered significant when a number of histogram counts within a significance-determination window centered at a peak location meets a first criterion.
26. The non-transitory computer-readable medium as described in claim 15 , wherein determining an additive noise estimate based on a comparison of the computed peak-width measure and a width threshold includes setting the additive noise estimate to a default value, if the computed peak-width measure is greater than the width threshold.
27. The non-transitory computer-readable medium as described in claim 26 , wherein said default value is a noise value associated with a previous frame in an image sequence associated with said image-channel image.
28. The non-transitory computer-readable medium as described in claim 15 , wherein determining an additive noise estimate based on a comparison of the computed peak-width measure and a width threshold includes computing an additive-noise estimate using a histogram-bin location of said first significant peak and said peak-width measure, if the computed peak-width measure is less than the width threshold.
29. A computing system a non-transitory computer readable storage medium having instructions stored thereon implementing a method for estimating additive noise in an image, said method comprising:
computing a statistical feature value for each image patch in a plurality of image patches associated with an image-channel image;
forming a histogram of said computed statistical feature values;
identifying a first significant peak in said histogram;
computing a peak-width measure of the identified first significant peak; and
determining an additive-noise estimate by setting an additive noise estimate to a default value, when the computed peak-width measure is greater than a width threshold and by computing an additive-noise estimate using a histogram-bin location of said first significant peak and said peak-width measure, when the computed peak-width measure is not greater than the width threshold.