Image content enhancement using a dictionary technique
A system for determining a high resolution image includes receiving a low resolution image and determining a vector of a patch of the low resolution image based upon a low resolution dictionary. The system includes determining a high resolution patch based upon a high resolution dictionary and the vector and determining the high resolution image based upon the high resolution patch.
1. A method for determining a high resolution image comprising:
(a) receiving a low resolution image;
(b) determining a vector for a patch of said low resolution image based upon a total number of non-zero dictionary elements of a low resolution dictionary that are used to said determine said vector, where said total number is modified using a parameter that balances the sparsity of a solution by adjusting the number of said elements used to determine said vector, said parameter inversely varied over said image relative to a standard deviation of a localized patch of said low resolution image;
(c) determining a high resolution patch based upon a high resolution dictionary and said vector;
(d) determining said high resolution image based upon said high resolution patch in a manner suitable to enhance the visual appearance of the high resolution image compared to said low resolution image and free from a constraint of reconstructing an original high resolution image corresponding to said low resolution image.
2. The method of claim 1 wherein said patch of said low resolution image for determining said vector is based upon a feature of said patch.
3. The method of claim 1 wherein said vector is a sparse vector containing a small number of non-zero weights.
4. The method of claim 1 wherein said low resolution dictionary and said high resolution dictionary are trained based upon a set of high resolution images.
5. The method of claim 1 wherein said determining said vector is based upon a variable factor dependent on the content of said low resolution image.
6. The method of claim 5 wherein said variable factor determines a number of non-zero weights in said vector.
7. The method of claim 2 wherein said feature includes at least one derivative of said low resolution image.
8. The method of 4 wherein said training is based upon a degradation operator.
9. The method of claim 8 wherein said degradation operator is a Gaussian blur.
10. The method of claim 8 wherein said degradation operator is introduction of compression artifacts.
11. The method of claim 8 wherein said degradation operator is loss of texture detail.
12. The method of claim 5 wherein said variable factor estimates localized texture content.
13. The method of claim 1 wherein said parameter is inversely varied in a stepwise manner over said image relative to a standard deviation of a localized patch of said low resolution image.
14. The method of claim 2 wherein said feature includes a Gabor filter.
15. The method of claim 2 wherein said feature includes a steerable complex pyramid.
16. The method of claim 2 wherein said feature includes a Sobel filter.
17. The method of claim 2 wherein said feature includes a Laplacian filter.
18. The method of claim 2 wherein said feature includes a 45 degree and 135 degree filter.
19. The method of claim 1 wherein said determining said high resolution patch is based upon a residue.
20. The method of claim 1 wherein said high resolution image is based upon degrading said low resolution image in a manner to determine dictionary based information.
21. The method of claim 1 wherein said high resolution patch is modified to selectively increase its intensity.
22. The method of claim 21 wherein said selectively increase is based upon image content.