Systems, methods, and computer program products for searching and sorting images by aesthetic quality

1. A method for assigning an aesthetic score to an image, comprising:

receiving an image comprising a set of global features;

extracting a set of global features for the image, the extraction comprising determining a plurality of scales, for each of the plurality of scales, segmenting the image into a plurality of regions, and for each of the plurality of scales and the plurality of regions:

computing a plurality of nearest and farthest regions according to a distance function;

computing a local histogram of oriented gradients; and

computing a color;

encoding the extracted set of global features into a high-dimensional feature vector;

reducing the dimension of the high-dimensional feature vector; and

applying a machine-learned model to assign an aesthetic score to the image, wherein a more aesthetically-pleasing image is given a higher aesthetic score and a less aesthetically-pleasing image is given a lower aesthetic score.

2. The method of claim 1 , wherein the segmentation comprises using a superpixel algorithm.

3. The method of claim 2 , wherein the superpixel algorithm further comprises using a simple linear iterative clustering (SLIC) algorithm.

4. The method of claim 1 , wherein the computing a plurality of nearest and farthest regions according to a distance function comprises at least one of a distance in CIE-LAB space and a distance in appearance space, wherein the appearance space comprises the local histogram of oriented gradients.

5. The method of claim 1 , wherein the high-dimensional feature vector is a Fisher vector.

6. The method of claim 1 , wherein the high-dimensional feature vector is a bag-of-visual-words vector.

7. The method of claim 1 , wherein the machine-learned model is learned using a LambdaMART algorithm.

8. The method of claim 1 , wherein the machine-learned model is learned using a LamdaRank algorithm.

9. The method of claim 1 , wherein the machine-learned model is learned using a RankNet algorithm.

10. The method of claim 1 , wherein the machine-learned model is learned using a RankSVM algorithm.

11. The method of claim 1 , wherein the machine-learned model is learned using a method that minimizes at least one of pairwise and list-wise ranking loss.

12. The method of claim 1 , wherein the reduction of the high-dimensional feature vector comprises applying a machine-learned projection matrix to keep the distance between pairs of aesthetic images and non-aesthetic images large and the distance between pairs of aesthetic images and aesthetic images and the distance between pairs of non-aesthetic images and non-aesthetic images small.

13. The method of claim 12 , wherein the machine-learned projection matrix is learned using a WSABIE algorithm.

14. The method of claim 1 , wherein the machine-learned model is trained on a subset of training data, whereby the assigned rank is personalized to the subset.

15. The method of claim 14 , wherein the subset of training data is based on a single user.

16. The method of claim 14 , wherein the subset of training data is based on a group of users.

17. The method of claim 14 , wherein the subset of training data is based on an image genre.

18. A method for ranking images by aesthetic quality for display, comprising:

receiving a plurality of images;

assigning an aesthetic score to each of the plurality of images, wherein the aesthetic score is generated by the method of claim 1 ; and

displaying the plurality of images in an order based on the aesthetic score.

19. The method of claim 18 , wherein the displaying the plurality of images comprises displaying a subset of the plurality of images.

20. The method of claim 19 , wherein the subset of the plurality of images is limited to a fixed number of images.

21. The method of claim 19 , wherein the subset of the plurality of images is based on a minimum aesthetic score.

22. A method for searching images by aesthetic quality, comprising:

receiving a search query;

searching a plurality of images based on the search query, whereby a list of search results is generated; and

displaying the list of search results by the method of claim 18 .

23. The method of claim 22 , wherein the searching a plurality of images comprises searching a plurality of metadata associated with the plurality of images.

24. A device for assigning an aesthetic score to an image, comprising:

a processor;

a memory coupled to the processor; and

a network interface coupled to the processor,

wherein the processor is configured to:

receive an image comprising a set of global features;

extract a set of global features for the image, the extraction comprising determining a plurality of scales, for each of the plurality of scales, segmenting the image into a plurality of regions, and for each of the plurality of scales and the plurality of regions:

computing a plurality of nearest and farthest regions according to a distance function;

computing a local histogram of oriented gradients; and

computing a color;

encode the extracted set of global features into a high-dimensional feature vector;

reduce the dimension of the high-dimensional feature vector; and

apply a machine-learned model to assign an aesthetic score to the image, wherein a more aesthetically-pleasing image is given a higher aesthetic score and a less aesthetically-pleasing image is given a lower aesthetic score.

25. A computer program product for assigning an aesthetic score to an image, said computer program product comprising a non-transitory computer readable medium storing computer readable program code embodied in the medium, said computer program product comprising:

program code for receiving an image comprising a set of global features;

program code for extracting a set of global features for the image, the extraction comprising determining a plurality of scales, for each of the plurality of scales, segmenting the image into a plurality of regions, and for each of the plurality of scales and the plurality of regions:

computing a plurality of nearest and farthest regions according to a distance function;

computing a local histogram of oriented gradients; and

computing a color;

program code for encoding the extracted set of global features into a high-dimensional feature vector;

program code for reducing the dimension of the high-dimensional feature vector; and

program code for applying a machine-learned model to assign an aesthetic score to the image, wherein a more aesthetically-pleasing image is given a higher aesthetic score and a less aesthetically-pleasing image is given a lower aesthetic score.