Method and system for learning-based quality assessment of images

• US 20060147107A1
• Filed: 01/04/2005
• Published: 07/06/2006
• Est. Priority Date: 01/04/2005
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Ma, Wei-Ying, Zhang, Hong-Jiang, Li, Ming Jing
• Granted Patent: US 7,545,985 B2
• Time in Patent Office: Days
• US Class Current: 382/159
• CPC Class Codes: G06T 7/0002 Inspection of images, e.g. flaw detection
• CPC Class Codes: G06V 10/993 Evaluation of the quality of the acquired pattern

Abstract

A method and system for learning-based assessment of the quality of an image is provided. An image quality assessment system trains an image classifier based on a training set of sample images that have quality ratings. To train the classifier, the assessment system generates a feature vector for each sample image representing various attributes of the image. The assessment system may train the classifier using an adaptive boosting technique to calculate a quality score for an image. Once the classifier is trained, the assessment system may calculate the quality of an image by generating a feature vector for that image and applying the trained classifier to the feature vector to calculate the quality score for the image.

Claims

1. A method in a computer system for assessing quality of an image, the method comprising:

providing a first set of images, each image having a quality rating;

training a classifier to indicate quality of images using the first set of images and their quality ratings; and

applying the trained classifier to assess the quality of the image.

2. The method of claim 1 wherein the classifier is an adaptive boosting classifier.

3. The method of claim 1 wherein the quality ratings are high-quality and low-quality.

4. The method of claim 1 wherein the trained classifier provides a quality score.

5. The method of claim 1 including generating a feature vector for each image of the first set wherein the classifier is trained using the generated feature vectors.

6. The method of claim 5 including generating a feature vector for the image wherein the trained classifier is applied to the generated feature vector.

7. The method of claim 5 wherein the features indicate a distortion evaluation.

8. The method of claim 5 wherein the features indicate a holistic quality evaluation.

9. The method of claim 8 wherein a feature is colorfulness.

10. The method of claim 8 wherein a feature is blurness.

11. The method of claim 8 wherein a feature is contrast.

12. The method of claim 8 wherein a feature is saliency.

13. The method of claim 1 including:

providing a second set of images;

providing a first quality score for each image of the second set;

generating a second quality score for each image of the second set by applying the trained classifier; and

generating a mapping function based on the first and second quality scores for the second set of images.

14. The method of claim 13 wherein the mapping function is represented as:

Ps(j)=α

+β

·

Qm(j)^γ

where Ps(j) is the quality score for image j, Qm(j) is the quality score generated by the classifier, and α

, β

, and γ

are mapping parameters.

15. The method of claim 14 wherein the mapping parameters are selected to minimize

$\mathrm{MSE}=\frac{1}{N_{\mathrm{aho}}}$

∑

j = 1 N aho ⁢

 

⁢

( Ps ⁢

 

⁢

( j ) - Mhs ⁢

 

⁢

( j ) ) 2

where Mhs(j) is the mean human observer score of image j and N_aho is the number of images used to determine the parameters.

16. The method of claim 1 wherein images produced by a professional have a quality rating of high and images of a non-professional have a quality rating of low.

17. A computer-readable medium containing instructions for controlling a computer system to assess quality of an image, by a method comprising:

providing a first set and a second set of images;

providing a quality rating for each image of the first set;

providing a first quality score for each image of the second set;

training a classifier to indicate quality of images using the first set of images and their quality ratings;

calculating a second quality score for each image of the second set using the trained classifier; and

generating a mapping function based on the trained classifiers and the first and second quality scores;

wherein a quality score for the image is calculated using the trained classifier and generated mapping function.

18. The computer-readable medium of claim 17 wherein the first and second sets of images are the same images.

19. The computer-readable medium of claim 17 wherein the classifier is an adaptive boosting classifier.

20. The computer-readable medium of claim 17 wherein the quality ratings are high-quality and low-quality.

21. The computer-readable medium of claim 17 including generating a feature vector for each image of the first set wherein the classifier is trained using the generated feature vectors.

22. The computer-readable medium of claim 21 wherein the features indicate a distortion evaluation.

23. The computer-readable medium of claim 21 wherein the features indicate a holistic quality evaluation.

24. The computer-readable medium of claim 17 wherein the mapping function is represented as:

Ps(j)=α

+β

·

Qm(j)^γ

where Ps(j) is the quality score for image j, Qm(j) is the quality score generated by the classifier, and α

, β

, and γ

are mapping parameters.

25. The computer-readable medium of claim 24 wherein the mapping parameters are selected to minimize

$\mathrm{MSE}=\frac{1}{N_{\mathrm{aho}}}$

∑

j = 1 N aho ⁢

 

⁢

( Ps ⁢

 

⁢

( j ) - Mhs ⁢

 

⁢

( j ) ) 2

where Mhs(j) is the mean human observer score of image j and N_aho is the number of images used to determine the parameters.

26. The computer-readable medium of claim 17 wherein images produced by a professional have a quality rating of high and images of a non-professional have a quality rating of low.

27. A computer system for assessing quality of an image, comprising:

a component that trains a classifier to indicate quality of images, the training based on a first set of images and their quality ratings; and

a component that calculates a quality score for the image by applying the trained classifier.

28. The system of claim 27 wherein the classifier is an adaptive boosting classifier.

29. The system of claim 27 wherein the quality ratings are high-quality and low-quality.

30. The system of claim 27 including a component that generates feature vectors for images, the feature vectors being used for training and calculating the quality score.

31. The system of claim 30 wherein the features indicate a distortion evaluation.

32. The system of claim 30 wherein the features indicate a holistic quality evaluation.

33. The system of claim 27 including a component that generates a mapping function based on first and second quality scores for a second set of images, the first quality scores indicating human evaluation of quality and the second quality scores being generated using the trained classifier.

34. The system of claim 33 wherein the mapping function is represented as:

Ps(j)=α

+β

·

Qm(j)^γ

where Ps(j) is the quality score for image j, Qm(j) is the quality score generated by the classifier, and α

, β

, and γ

are mapping parameters.

35. The system of claim 34 wherein the mapping parameters are selected to minimize

$\mathrm{MSE}=\frac{1}{N_{\mathrm{aho}}}$

∑

j = 1 N aho ⁢

 

⁢

( Ps ⁢

 

⁢

( j ) - Mhs ⁢

 

⁢

( j ) ) 2

where Mhs(j) is the mean human observer score of image j and N_aho is the number of images used to determine the parameters.

36. The system of claim 27 wherein images produced by a professional have a quality rating of high and images of a non-professional have a quality rating of low.