Image cropping using supervised learning

1. A method for generating thumbnail images, comprising the operations of:

extracting a plurality of pixel-level features from each of a plurality of source images, wherein each of the source images is associated with a thumbnail created by an editor;

generating a plurality of unique bounding boxes for each source image;

calculating a plurality of region-level features for each bounding box, wherein each region-level feature results from the aggregation of pixel values for one of the pixel-level features;

learning a regression model, using at least the calculated region-level features and the thumbnail associated with the source image; and

choosing a thumbnail from a plurality of unique bounding boxes in a new image, based at least in part on application of the regression model, wherein application of the regression model occurs if the source image is not approximately square, and wherein each of the operations is executed by one or more processors.

2. The method of claim 1 , wherein a center-fit approach is applied to the source image to obtain a thumbnail, if the source image is approximately square.

3. The method of claim 1 , wherein learning the regression model includes measuring a similarity between a bounding box in a source image and the thumbnail for the source image.

4. The method of claim 1 , wherein the operation of choosing a thumbnail from a plurality of unique bounding boxes in a new image includes generating a plurality of bounding boxes for the new image and calculating a plurality of region-level features from pixel-level features for each bounding box.

5. The method of claim 1 , further comprising the operation of using a thumbnail received from an editor instead of the chosen thumbnail, if the chosen thumbnail is of insufficient quality as measured against a scoring threshold.

6. The method of claim 5 , wherein the regression model is updated with the received thumbnail.

7. The method of claim 1 , wherein the regression model includes a support vector regression machine.

8. The method of claim 1 , wherein at least one of the pixel-level features is based on a saliency map.

9. A computer-readable storage medium, which is non-transitory, storing a program, wherein the program, when executed, instructs a processor to perform the following operations:

extract a plurality of pixel-level features from each of a plurality of source images, wherein each of the source images is associated with a thumbnail created by an editor;

generating a plurality of unique bounding boxes for each source image;

calculate a plurality of region-level features for each bounding box, wherein each region-level feature results from the aggregation of pixel values for one of the pixel-level features;

learn a regression model, using at least the calculated region-level features and the thumbnail associated with the source image; and

choose a thumbnail from a plurality of unique bounding boxes in a new image, based at least in part on application of the regression model, wherein application of the regression model occurs if the source image is not approximately square.

10. The computer-readable storage medium of claim 9 , wherein a center-fit approach is applied to the source image to obtain a thumbnail, if the source image is approximately square.

11. The computer-readable storage medium of claim 9 , wherein learning the regression model includes measuring a similarity between a bounding box in a source image and the thumbnail for the source image.

12. The computer-readable storage medium of claim 9 , wherein choosing a thumbnail from a plurality of unique bounding boxes in a new image includes generating a plurality of bounding boxes for the new image and calculating a plurality of region-level features from pixel-level features for each bounding box.

13. The computer-readable storage medium of claim 9 , further comprising the operation of using a thumbnail received from an editor instead of the chosen thumbnail, if the chosen thumbnail is of insufficient quality as measured against a scoring threshold.

14. The computer-readable storage medium of claim 13 , wherein the regression model is updated with the received thumbnail.

15. The computer-readable storage medium of claim 9 , wherein the regression model includes a support vector regression machine.

16. The computer-readable storage medium of claim 9 , wherein at least one of the pixel-level features is based on a saliency map.

17. A method for generating thumbnail images, comprising the operations of:

extracting a plurality of disaggregated features from each of a plurality of source images, wherein each of the source images is associated with a thumbnail created by an editor;

generating a plurality of unique bounding boxes for each source image;

calculate a plurality of region-level features for each bounding box, wherein each of these region-level feature results from the aggregation of values for one of the disaggregated features;

determining another region-level feature for each bounding box;

learning a regression model, using at least the calculated region-level features, the other region-level feature, and the thumbnail associated with the source image;

choosing a thumbnail from a plurality of candidate regions in a new image, based at least in part on application of the regression model; and

using a thumbnail received from an editor instead of the chosen thumbnail, if the chosen thumbnail is of insufficient quality as measured against a scoring threshold, wherein the regression model is updated with the received thumbnail and wherein each of the operations is executed by one or more processors.

18. The method of claim 17 , wherein the other region-level feature is based on a measure of brightness.

19. The method of claim 17 , wherein application of the regression model occurs if the source image is not approximately square.

20. The method of claim 19 , wherein a center-fit approach is applied to the source image to obtain a thumbnail, if the source image is approximately square.