Soft edge smoothness prior and application on alpha channel super resolution
View Patent ↗Systems and methods are disclosed for processing a low resolution image by performing a high resolution edge segment extraction on the low resolution image; performing an image super resolution on each edge segment; performing reconstruction constraint reinforcement; and generating a high quality image from the low quality image.
1. A method to process a low resolution image, comprising:
a. capturing the low resolution image from an image sensor:
b. performing a high resolution edge segment extraction on the low resolution image;
c. performing an image super resolution on each edge segment;
d. performing reconstruction constraint reinforcement;
e. generating a high quality image from the low quality image; and
f. rendering the high quality image on paper or a screen.
2. The method of claim 1 , comprising performing alpha matting to get alpha channels and colors on each edge segment.
3. The method of claim 1 , comprising performing bicubic interpolation on each edge segment.
4. The method of claim 1 , comprising applying graph cuts on the bicubic interpolated data to generate a super resolution alpha channel.
5. The method of claim 4 , comprising assigning one or more colors to the super resolution alpha channel.
6. The method of claim 1 , comprising deriving a smooth edge prior for the low resolution image.
7. The method of claim 6 , wherein the smooth edge prior comprises I=α×F+(1−α)×B.
8. The method of claim 1 , wherein the high resolution edge segment extraction comprises using one or more different size neighborhood.
9. The method of claim 1 , comprising getting different distance maps.
10. The method of claim 1 , comprising applying the Geocuts method.
11. The method of claim 1 , comprising applying a defined soft cut metric for intensity image.
12. The method of claim 1 , comprising applying an alpha matting technique to provide a soft edge smoothness prior on natural color images.
13. The method of claim 1 , comprising measuring a soft edge smoothness by approximating an average length of one or more level lines.
14. The method of claim 13 , comprising adding the soft edge smoothness as a prior term for super resolution to achieve edge preservation and edge smoothness.
15. The method of claim 1 , comprising transforming color image super resolution into an alpha channel super resolution and alpha matting.
16. The method of claim 1 , comprising using a closed form alpha matting solution to describe each edge segment in a unified way through the alpha channel.
17. The method of claim 16 , wherein color information from a plurality of channels is utilized simultaneously.
18. The method of claim 1 , comprising performing edge segment extraction and region assignment to get {c i } and {P i }.
19. The method of claim 18 , for each segment c i , processing P i to:
compute l , l and α l from I l by a closed form alpha matting solution,
perform alpha channel SR to get α h from α l by single channel super resolution with soft edge smoothness prior; and
synthesize a high resolution patch by l , l and α h .
20. The method of claim 19 , reinforcing a reconstruction constraint for an image by back-projection.
21. The method of claim 1 , comprising applying a canny edge detection method to extract continues edges.
22. The method of claim 1 , comprising applying a corner detection method based on curvature scale space is applied.
23. The method of claim 22 , comprising breaking one or more edges into segments.