Automatic image adjustment parameter correction

1. A method, comprising:

receiving a plurality of image pairs, each of the plurality of image pairs being identified by a category and including a raw image and an adjusted version of the raw image;

generating automatically predicted adjustment parameters for each of the plurality of image pairs by applying a regression algorithm to each image pair, the automatically predicted adjustment parameters comprising a plurality of curves that each relate to one or more parameters of an image pair;

selecting a plurality of the automatically predicted adjustment parameters generated from image pairs identified by the same category;

modifying one or more of the selected automatically predicted adjustment parameters based on a rule that defines a minimum or maximum slope, the automatically predicted adjustment parameters defining a global adjustment to an image and are based on machine learning prediction; and

globally adjusting a new image based on the modified automatically predicted adjustment parameters.

2. The method of claim 1 , wherein the rule includes a bound for the automatically predicted adjustment parameters.

3. The method of claim 1 , wherein the one or more automatically predicted adjustment parameters is a tone curve.

4. The method of claim 1 , further comprising prior to performing said modifying, receiving the automatically predicted adjustment parameters.

5. The method of claim 1 , further comprising prior to performing said modifying, generating the automatically predicted adjustment parameters, wherein said generating includes performing the machine learning prediction.

6. The method of claim 1 , wherein said generating comprises training a regression algorithm, wherein the plurality of curves are the automatically predicted adjustment parameters;

wherein said modifying includes modifying at least one of the plurality of curves based on the rule; and

wherein said globally adjusting the new image is based on the at least one modified curve.

7. The method of claim 6 , wherein said globally adjusting the new image includes the trained regression algorithm performing a weighted combination of the plurality of curves including the at least one modified curve and applying the weighted combination to the new image.

8. The method of claim 1 , wherein said globally adjusting the new image includes globally adjusting one or more tonal parameters of the new image.

9. The method of claim 1 , wherein the machine learning prediction is a supervised learning prediction.

10. A non-transitory computer-readable storage device comprising program instructions stored thereon, the program instructions are computer-executable to implement operations comprising:

receiving a plurality of image pairs, each image pair of the plurality of image pairs including a raw image and an adjusted version of the raw image;

generating an automatically predicted adjustment parameter from the plurality of image pairs by applying a regression algorithm to each image pair, the automatically predicted adjustment parameter comprising a plurality of curves that relates to one or more parameters of an image pair;

receiving adjustments to a subset of image pairs of the plurality of image pairs;

computing a transferred adjustment by applying the regression algorithm to the received adjustments;

modifying the automatically predicted adjustment based the computed transferred adjustment, the automatically predicted adjustment describes a global adjustment to an image and is based on a machine learning prediction; and

globally adjusting a new image based on the modified automatically predicted adjustment.

11. The non-transitory computer-readable storage medium of claim 10 , wherein the computed transferred adjustment includes a bound for the automatically predicted adjustment.

12. The non-transitory computer-readable storage medium of claim 10 , wherein the automatically predicted adjustment is a tone curve.

13. The non-transitory computer-readable storage medium of claim 10 , wherein the program instructions are further computer-executable to implement:

prior to performing said modifying, generating the automatically predicted adjustment, wherein said generating includes performing the machine learning prediction.

14. The non-transitory computer-readable storage medium of claim 10 , wherein said generating comprises training a regression algorithm, wherein the plurality of curves is the automatically predicted adjustment;

wherein said modifying includes modifying at least one of the plurality of curves based on the rule; and

wherein said globally adjusting the new image is based on the plurality of curves including the at least one modified curve.

15. The non-transitory computer-readable storage medium of claim 14 , wherein said globally adjusting the new image includes the trained regression algorithm performing a weighted combination of the plurality of curves including the at least one modified curve and applying the weighted combination to the new image.

16. The non-transitory computer-readable storage medium of claim 10 , wherein said globally adjusting the new image includes globally adjusting one or more tonal parameters of the new image.

17. A system, comprising:

at least one processor; and

a memory comprising program instructions, wherein the program instructions are executable by the at least one processor to:

modify an automatically predicted global adjustment based on a rule that includes a bound for the automatically predicted global adjustment determined by features of one or more local regions of an image, the automatically predicted global adjustment describes a global adjustment to the image that treats every pixel of the image the same and is based on a machine learning prediction; and

globally adjust a new image based on the modified automatically predicted global adjustment.

18. The system of claim 17 , wherein the automatically predicted global adjustment is a tone curve.

19. The system of claim 17 , wherein the program instructions are further executable by the at least one processor to:

receive a plurality of image pairs, wherein each image pair of the plurality of image pairs includes a raw image and an adjusted version of the raw image; and

generate a plurality of curves that each relate one or more parameters of the raw image to a corresponding adjusted image of an image pair, wherein said generating comprises training a regression algorithm, wherein the plurality of curves is the automatically predicted global adjustment;

wherein said modifying includes modifying at least one of the plurality of curves based on the rule;

wherein said globally adjusting the new image is based on the plurality of curves including the at least one modified curve.

20. The system of claim 17 , wherein said globally adjusting the new image includes globally adjusting one or more tonal parameters of the new image.

21. The system of claim 17 , wherein the features of one or more local regions of the image comprise one or more of detected faces, local histograms, color distributions, or scene descriptors.