computerized aesthetic judgment of images

• US 6,816,847 B1
• Filed: 09/23/1999
• Issued: 11/09/2004
• Est. Priority Date: 09/23/1999
• Status: Expired due to Term

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Toyama, Kentaro
• Primary Examiner(s): Knight, Anthony
• Assistant Examiner(s): Booker, Kelvin
• Application Number:
  US09/404,940
• Time in Patent Office: 1,874 Days
• Field of Search: 706/14, 706/20, 706/18, 382/157, 382/158, 382/159, 382/156, 382/224, 382/128, 382/133, 382/155, 382/185
• US Class Current: 706/14
• CPC Class Codes: G06F 18/00 Pattern recognition
• CPC Class Codes: G06T 2207/30168 Image quality inspection
• CPC Class Codes: G06T 7/0002 Inspection of images, e.g. flaw detection

Abstract

Computerized aesthetic judgment of images is disclosed. In one embodiment, a computer-implemented method inputs a training set of images, where each image has a corresponding set of aesthetic scores. The method trains a classifier based on the training set, and outputs the classifier. An image can then be input into the classifier, such that an aesthetic score for the image is generated by the classifier and output. Furthermore, recommendations can be generated to improve the aesthetic score for the image, which are also output.

Claims

1. A computer-implemented method comprising:

inputting a training set including a plurality of images and a corresponding plurality of aesthetic scores for the images;

training a classifier to provide aesthetic scores based on the training set;

generating an aesthetic score for an image based on the classifier, and

generating a recommendation that suggests how to improve the aesthetic score for the image by manipulating visual elements of the image.

2. The method of claim 1, wherein generating a recommendation comprises utilizing a gradient ascent.

3. The method of claim 1, wherein generating a recommendation comprises performing a local search.

4. The method of claim 1, wherein training a classifier comprises training one of a Bayesian classifier, a Support Vector Machine (SVM) classifier, a neural net classifier, and a decision tree classifier.

5. The method of claim 1, wherein training a classifier comprises utilizing feature selection to correlate at least one image feature of the images with their corresponding aesthetic scores.

6. The method of claim 5, wherein utilizing feature selection to correlate at least one image feature comprises utilizing feature selection to correlate at least one image feature selected from the group essentially consisting of:

color presence, color distribution, geometrical quantities of segmented image parts, coefficients of image transformations, and higher-level image representations.

18. The method of claim 1, wherein training a classifier comprises training one of a Bayesian classifier, a Support Vector Machine (SVM) classifier, a neural net classifier, and a decision tree classifier.

7. A computer-implemented method comprising:

inputting an image;

generating an aesthetic score for the image by utilizing a classifier previously trained on a training set including a plurality of images and a corresponding plurality of aesthetic scores for the images; and

,

generating a recommendation to improve the aesthetic score.

8. The method of claim 7, wherein generating an aesthetic score comprises generating an aesthetic score based on at least one image feature of the image.

9. The method of claim 8, wherein generating an aesthetic score based on at least one image feature of the image comprises generating an aesthetic score based on at least one image feature selected form the group essentially consisting of:

color presence, color distribution, geometrical quantifies of segmented image parts, coefficients of image transformations, and higher-level image representations.

10. The method claim 7, wherein utilizing a classifier comprises utilizing one of a Bayesian classifier, a Support Vector Machine (SVM classifier, a neural net classifier, and a decision tee classifier.

11. A computer-implemented method comprising:

inputting an image and a corresponding aesthetic score for the image previously generated by utilizing a classifier previously trained on a training set including a plurality of images and a corresponding plurality of aesthetic scores for the images;

generating a recommendation to improve the aesthetic score for the image; and

,

outputting the recommendation.

12. The method of claim 11, wherein generating a recommendation comprises utilizing a gradient ascent.

13. The method of claim 11, wherein generating a recommendation comprises performing a local search.

14. The method of claim 11, wherein inputting an image and a corresponding aesthetic score comprises inputting an image and a corresponding aesthetic score for the image previously generated based on at least one image feature of the image.

15. The method of claim 14, wherein inputting an image and a corresponding aesthetic score for the image previously generated based on at least one image feature of the image comprises inputting an image and a corresponding aesthetic score for the image previously generated based on at least one image feature of the image selected from the group essentially consisting of:

color presence, color distribution, geometrical quantifies of segmented image parts, coefficients of image transformations, and higher-level image representations.

16. The method of claim 11, wherein inputting an image and a corresponding aesthetic score comprises inputting an image and a corresponding aesthetic score for the image previously generated by utilizing one of a Bayesian classifier, a Support Vector Machine (SVM) classifier, a neural net classifier, and a decision tree classifier.

17. A machine-readable medium having instructions stored thereon for execution by a processor to perform a method comprising:

inputting a training set including a plurality of images and a corresponding plurality of aesthetic scores for the images;

training a classifier to provide aesthetic scores based on the training set;

inputting an image and utilizing the classifier to generate a corresponding aesthetic score for the image; and

generating a recommendation that suggests which visual elements of the image to manipulate to improve the corresponding aesthetic score, wherein the visual elements are capable of being manipulated by a computer.

19. The method of claim 17, wherein training a classifier comprises utilizing feature selection to correlate at least one image feature of the images with their corresponding aesthetic scores.

20. The method of claim 17, wherein the image includes at least one of;

(a) a web page;

(b) a digital photograph;

(c) an electronic greeting card;

(d) a desktop-publishing document;

(e) a word-processing document.

21. A machine-readable medium having instructions stored thereon for execution by a processor to perform a method comprising:

inputting an image;

generating an aesthetic score for the image by utilizing a classifier previously trained on a training set including a plurality of images and a corresponding plurality of aesthetic scores for the images;

outputting the image; and

,

generating and outputting a recommendation as to how to improve the aesthetic score for the image.

22. The medium of claim 21, wherein generating an aesthetic score comprises generating an aesthetic score based on at least one image feature of the image.

23. The medium of claim 21, wherein utilizing a classifier comprises utilizing one of a Bayesian classifier, a Support Vector Machine (SVM) classifier, a neural net classifier, and a decision tree classifier.

24. A machine-readable medium having instructions stored thereon for execution by a processor to perform a method comprising:

inputting an image and a corresponding aesthetic score for the image previously generated by utilizing a classifier previously trained on a training set including a plurality of images and a corresponding plurality of aesthetic scores for the images;

generating a recommendation to improve the aesthetic score for the image; and

,

outputting the recommendation.

25. The medium of claim 24, wherein generating a recommendation comprises utilizing a gradient ascent.

26. The medium of claim 24, wherein generating a recommendation comprises performing a local search.

27. The medium of claim 24, wherein inputting an image and a corresponding aesthetic score comprises inputting an image and a corresponding aesthetic score for the image previously generated based on at least one image feature of the image.

28. The medium of claim 24, wherein inputting an image and a corresponding aesthetic score comprises inputting an image and a corresponding aesthetic score for the image previously generated by utilizing one of a Bayesian classifier, a Support Vector Machine (SVM) classifier, a neural net classifier, and a decision tree classifier.