Method and system for learning-based quality assessment of images

• US 7,545,985 B2
• Filed: 01/04/2005
• Issued: 06/09/2009
• Est. Priority Date: 01/04/2005
• Status: Expired due to Fees

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Ma, Wei-Ying, Zhang, Hong-Jiang, Li, Ming Jing
• Primary Examiner(s): Bella; Matthew C
• Assistant Examiner(s): Rahmjoo; Mike
• Application Number:
  US11/029,913
• Publication Number:
  US 20060147107A1
• Time in Patent Office: 1,617 Days
• Field of Search: 382/224, 382/227
• US Class Current: 382/224
• 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 computer-readable medium containing instructions for controlling a computer system with a processor and a memory to assess quality of an image, the instructions implementing steps comprising:

accessing a first set and a second set of images;

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

accessing 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 classifier and the first and second quality scores;

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

wherein the classifier is an adaptive boosting classifier, and

wherein the instructions are executed by the processor after being loaded into memory from the computer-readable medium.

2. The computer-readable medium of claim 1 wherein the first and second sets of images are the same images.

3. The computer-readable medium of claim 1 wherein the quality ratings are high-quality and low-quality.

4. The computer-readable medium 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.

5. The computer-readable medium of claim 4 wherein the features indicate a distortion evaluation.

6. The computer-readable medium of claim 4 wherein the features indicate a holistic quality evaluation.

7. The computer-readable medium of claim 1 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.

8. The computer-readable medium of claim 7 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.

9. The computer-readable medium 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.