Method and system for generating a quality prediction table for quality-aware transcoding of digital images

1. A system for generating a quality metric prediction table for predicting quality in image transcoding, the system including:

a computer, comprising a processor, and a non-transitory computer readable storage medium having computer readable instructions stored thereon for execution by the processor, which cause the processor to:

determine an input encoding quality factor for each of a plurality of input images in a set of training images;

transcode the plurality of input images into corresponding output images, with corresponding output encoding quality factors and resolution scaling factors;

to compute a quality metric for each transcoding of each of the input images, comprising comparing each output image with the corresponding input image, the quality metric being a measure of distortion introduced by said each transcoding; and to store the computed quality metrics in a form of the quality metric prediction table, the quality metric prediction table having storage locations for the computed quality metrics, the storage locations being indexed by input encoding quality factor, output encoding quality factor and resolution scaling factor.

2. The system of claim 1 , wherein the quality metric prediction table is further indexed by a viewing condition, the viewing condition determining viewing scaling factors used in the comparing of each output image with the corresponding input image.

3. The system of the claim 1 , wherein the quality metric is based on Peak Signal to Noise Ratio (PSNR) of the output image compared with the input image.

4. The system of the claim 1 , wherein the quality metric is based on Maximum-Difference (MD) measure of the output image compared with the input image.

5. The system of the claim 1 , wherein the resolution scaling factor is incremented from 10% to 100% in steps of 10%.

6. The system of the claim 1 , wherein the output encoding quality factor is incremented from 10 to 100 in steps of 10.

7. The system of the claim 1 , wherein the input and output images are JPEG images.

8. The system of claim 1 , wherein the computer readable instructions further cause the processor to update the quality metrics stored in the quality metric prediction table with quality metrics computed by transcoding additional input images.

9. The system of claim 2 , wherein the viewing condition is chosen to optimize an image quality experienced by a viewer of the output image.

10. The system of claim 8 , wherein the computer readable instructions further cause the processor to:

generate LMS (least mean squares) estimators of the quality metrics computed for different training images, for corresponding combinations of the transcoding parameters; and

storing said estimators as the quality metrics at the storage locations indexed by the corresponding combinations of the transcoding parameters.

11. A method for generating a quality metric prediction table for predicting quality in image transcoding, the method comprising:

employing at least one hardware processor for:

(a) determining an input encoding quality factor for each of a plurality of input images in a set of training images;

(b) transcoding the plurality of input images into corresponding output images, with corresponding output encoding quality factors and resolution scaling factors;

(c) computing a quality metric for each transcoding of each of the input images, comprising comparing each output image with the corresponding input image, the quality metric being a measure of distortion introduced by said each transcoding;

(d) storing the computed quality metrics in a form of the quality metric prediction table, the quality metric prediction table having storage locations for the computed quality metrics, the storage locations being input encoding quality factor, output encoding quality factor and resolution scaling factor.

12. The method of claim 11 , further comprising indexing the quality metric prediction table by a viewing condition, the viewing condition determining viewing scaling factors used in the comparing of each output image with the corresponding input image.

13. The method of the claim 11 , wherein the step (c) further includes determining the quality metric based on Peak Signal to Noise Ratio (PSNR) of the output image compared with the input image.

14. The method of the claim 11 , wherein the step (c) further includes determining the quality metric based on Maximum Difference (MD) measure of the output image compared with the input image.

15. The method of the claim 11 , wherein the step (b) includes incrementing the resolution scaling factor from 10% to 100% in steps of 10%.

16. The method of the claim 11 , wherein the step (b) includes incrementing the output encoding quality factor from 10 to 100 in steps of 10.

17. The method of the claim 11 , wherein the input and output images are JPEG images.

18. The method of claim 11 , further comprising generating intermediate entries in the quality metric prediction table by interpolating between entries of the quality metric prediction table.

19. The method of claim 11 , further comprising updating the quality metrics stored in the quality metric prediction table with quality metrics computed by transcoding additional training images.

20. The method of claim 12 , further comprising choosing the viewing condition to optimize an image quality experienced by a viewer of the output image.

21. The method of claim 19 , further comprising generating LMS (least mean squares) estimators of the quality metrics computed for different training images, for corresponding combinations of the transcoding parameters, and storing said estimators as the quality metrics at the storage locations indexed by the corresponding combinations of the transcoding parameters.