Method and apparatus for measuring video quality using at least one semi-supervised learning regressor for mean observer score prediction

• US 8,824,783 B2
• Filed: 04/30/2010
• Issued: 09/02/2014
• Est. Priority Date: 04/30/2010
• Status: Expired due to Fees

• Current Assignee: Thomson Licensing (Vantiva SA)
• Original Assignee: Thomson Licensing (Vantiva SA)
• Inventors: Xu, Feng, Chen, Zhibo, Liu, Debing, Gu, Xiaodong
• Primary Examiner(s): SHAH, UTPAL D
• Application Number:
  US13/695,060
• Publication Number:
  US 20130050503A1
• Time in Patent Office: 1,586 Days
• Field of Search: 382/155, 382/159, 382/190
• US Class Current: 382/159
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06T 2207/10016 Video; Image sequence
• CPC Class Codes: G06T 2207/20081 Training; Learning
• CPC Class Codes: G06T 2207/30168 Image quality inspection
• CPC Class Codes: G06T 7/0002 Inspection of images, e.g. flaw detection
• CPC Class Codes: H04N 17/004 for digital television systems

Abstract

The invention is made in the technical field of video quality measurement. More precisely, the invention is related to mean observer score prediction using a trained semi-supervised learning regressor. That is, a method and apparatus for measuring video quality using a semi-supervised learning system for mean observer score prediction is proposed. Said semi-supervised learning system comprises at least one semi-supervised learning regressor and said method comprises training the learning system and retraining the trained learning system using a selection of test data wherein the test data is used for determining at least one mean observer score prediction using the trained learning system and the selection is indicated by a feedback received through a user interface upon presenting, in the user interface, said at least one mean observer score prediction. Doing so, prediction quality can be improved after re-training at least for the selection.

Claims

1. Method for measuring video quality using a semi-supervised learning system for mean observer score prediction, said semi-supervised learning system comprising at least one semi-supervised learning regressor, said method comprising training the learning system and retraining the trained learning system using a selection of test data wherein the test data is used for determining at least one mean observer score prediction using the trained learning system and the selection is indicated by a feedback received through a user interface upon presenting, in the user interface, said at least one mean observer score prediction.

2. Method according to claim 1, wherein said method further comprises receiving a label through said interface and using said label for labelling at least a part of the selection for re-training.

3. Method according to claim 1, said method comprises using training data used for training the learning system for re-training, also, wherein contribution of the selection and the training data for re-training is controlled by a first weight factor assigned to the selection and at least a different second weight factor assigned to the training data.

4. Method according to claim 3, wherein, at least for re-training, the training data comprises labelled and unlabelled training data, said at least a different second weight factor being assigned to the labelled training data and at least a different third weight factor being assigned to the unlabelled training data.

5. Method according to claim 4, further comprising determining a distance between two video frames, each of the two video frames being comprised in either the training data or the test data wherein the distance is determined using a distance metrics and at least one of the first weight factor and the at least a different second weight factor wherein the first weight factor is used, if at least one of the two video frames is comprised in the test data, the at least a different second weight factor is used, if at least one of the two video frames is comprised in the training data, and both, the first and the at least a different second weight factor are used for determining the distance, if one of the two video frames is comprised in the training data and the other is comprised in the test data.

6. Method according to claim 4, further comprising determining an observer score estimate for a data instance using a first neighbourhood of k nearest neighbours in the test data and at least a second neighbourhood of k nearest neighbours in the training data, wherein the observer score estimate is determined using the first weight factor and the at least a different second weight factor.

7. Method according to claim 4, wherein said semi-supervised learning system comprises two semi-supervised learning regressors, said method comprising training and retraining the two regressors wherein at least a part of the labelled training data is labelled by one of the two regressors, said at least a part of the labelled training data being used for re-training of the other of the two regressors.

8. Method according to claim 3, wherein at least one of the first and the at least a different second weight factor is received through said user interface.

9. Apparatus for video quality measurement, said apparatus comprising:

a semi-supervised learning system for mean observer score prediction, said semi-supervised learning system comprising at least one semi-supervised learning regressor,

a user interface for receiving a selection of test data upon presenting at least one mean observer score prediction, said at least one mean observer score prediction being determinable using test data and the learning system, wherein

the semi-supervised learning system is re-trainable using said selection of the test data.

10. The Apparatus of claim 9 adapted for receiving, through the user interface, a label for labelling at least a part of the selection for re-training.

11. The Apparatus of claim 9 adapted for retraining the learning system using said selection together with training data, said training data being used for training the learning system, also, wherein contribution of the selection and the training data for re-training is controlled by weight factors.

12. Apparatus according to claim 11, wherein, at least for re-training, the training data comprises labelled and unlabelled training data, wherein contribution of the selection and the training data for re-training is controlled by a first weight factor assigned to the selection and at least a different second weight factor assigned to the training data.

13. Apparatus according to claim 12, wherein, at least for re-training, the training data comprises labelled and unlabelled training data, said at least a different second weight factor being assigned to the labelled training data and at least a different third weight factor being assigned to the unlabelled training data.

14. Apparatus according to claim 13, further comprising first means for determining a distance between two video frames, each of the two video frames being comprised in either the training data or the test data, wherein said first means are adapted for determining the distance using a distance metrics and at least one of:

the first weight factor and the at least a different second weight factor wherein the first weight factor is used, if at least one of the two video frames is comprised in the test data, the at least a different second weight factor is used, if at least one of the two video frames is comprised in the training data, and both, the first and the at least a different second weight factor are used for determining the distance, if one of the two video frames is comprised in the training data and the other is comprised in the test data.

15. Apparatus according to claim 13, further comprising second means for determining an observer score estimate for a data instance using a first neighbourhood of k nearest neighbours in the test data and at least a second neighbourhood of k nearest neighbours in the training data, wherein said second means are adapted for determining the observer score estimate using the first weight factor and the at least a different second weight factor.

16. Apparatus according to claim 12, wherein said semi-supervised learning system comprises two semi-supervised learning regressors, said apparatus being adapted for training and retraining the two regressors wherein at least a part of the labelled training data is labelled by one of the two regressors, said at least a part of the labelled training data being used for re-training of the other of the two regressors.

17. Apparatus according to claim 12, wherein said user interface is further adapted for receiving, at least one of the first and the at least a different second weight factor.