System and method for quality-aware selection of parameters in transcoding of digital images
Several quality-aware transcoding systems and methods are described, in which the impact of both quality factor (QF) and scaling parameter choices on the quality of transcoded images are considered in combination. A basic transcoding system is enhanced by the addition of a quality prediction look-up table, and a method of generating such a table is also shown.
1. A method for selecting transcoding parameters and transcoding an input image into an output image for a terminal having terminal constraints including a maximum device file size, the method comprising:
(a) determining a file size, and an encoding quality factor QF(I) of the input image;
(b) providing an array of relative file size prediction values and predicted quality metric values characterizing a measure of distortion of an image introduced by transcoding, the array being indexed by encoding quality factor QF(I), output encoding quality factor QFT and a resolution scaling factor zT;
(c) computing a maximum relative file size as the maximum device file size divided by the file size of the input image;
(d) selecting, from the array, a set of feasible transcoding parameters, yielding those predicted relative file size values not exceeding the maximum relative file size, which correspond to the file size and the encoding quality factor QF(I) of the input image;
(e) obtaining, from the array, predicted quality metric values corresponding to the set of feasible transcoding parameters; and
(f) selecting those transcoding parameters from the set of feasible transcoding parameters, which yield the highest predicted quality metric value.
2. The method of claim 1 , further comprising transcoding the input image into the output image using the selected transcoding parameters corresponding to the highest predicted quality metric.
3. The method of claim 1 , wherein the array has been obtained from transcoding of a set of training images.
4. The method of claim 1 , wherein the set of feasible transcoding parameters comprises the resolution scaling factor zT not exceeding a maximum scaling factor z_max determined using predetermined maximum image dimensions of the output image and image dimensions of the input image.
5. The method of claim 1 , wherein the step (d) comprises interpolating between entries of the data structure for at least one of the indices.
6. The method of claim 1 , the step (d) further comprising limiting the set of feasible transcoding parameters to those combinations of the output encoding quality factor QFT and the resolution scaling factor zT yielding predicted relative file size values not exceeding the maximum relative file size.
7. The method of claim 1 , wherein the array comprises a multi-dimensional relative file size prediction table indexed by the QF(I), QFT and zT.
8. The method of claim 1 , wherein the array comprises a multi-dimensional quality prediction table indexed by the QF(I), QFT and zT.
9. The method of claim 1 , wherein the input image and the output image are JPEG images.
10. The method of claim 1 , wherein the quality metric values are based on the Maximum Difference (MD) measure of respective transcoded training images compared with corresponding training images in the set of training images.
11. The method of claim 1 , the step (e) further comprising:
sorting the set of feasible transcoding parameters according to the corresponding predicted quality metric values; and
truncating the set of feasible transcoding parameters by reducing it to a promising subset corresponding to high predicted quality metric values.
12. The method of claim 3 , wherein the set of training images comprises a first set of training images and a second set of training images, the method further comprising:
obtaining the predicted relative file size values from transcoding of the first set of training images; and
obtaining the predicted quality metric values from transcoding of the second set of training images.
13. The method of claim 3 , wherein the quality metric values are determined based on the Peak Signal to Noise Ratio (PSNR) measure of respective transcoded training images compared with corresponding training images in the set of training images.
14. The method of claim 6 , the step (d) further comprising limiting the set of feasible transcoding parameters to those combinations of zT and QFT wherein the resolution scaling factor zT does not exceed a maximum scaling factor z_max determined from predetermined maximum image dimensions of the output image and image dimensions of the input image.
15. The method of claim 14 , wherein the maximum scaling factor z_max is further defined according to a viewing scaling factor zV chosen based on anticipated viewing conditions of the output image.
16. A system for selecting transcoding parameters and transcoding an input image into an output image for a terminal having terminal constraints including a maximum device file size, the system comprising:
a processor; and
a computer readable storage medium having computer readable instructions stored thereon for execution for the processor, forming:
an image feature extraction module, determining a file size, and an encoding quality factor QF(I) of the input image;
an array of relative file size prediction values and predicted quality metric values characterizing a measure of distortion of an image introduced by transcoding, the array being indexed by encoding quality factor QF(I), output encoding quality factor QFT and a resolution scaling factor zT;
a quality and file size prediction module being configured to:
compute a maximum relative file size as the maximum device file size divided by the file size of the input image;
select from the array, a set of feasible transcoding parameters, yielding those predicted relative file size values not exceeding the maximum relative file size, which correspond to the file size and the encoding quality factor QF(I) of the input image; and
obtain, from the array, predicted quality metric values corresponding to the set of feasible transcoding parameters;
and
a quality-aware parameter selection module, selecting those transcoding parameters from the set of feasible transcoding parameters, which yield the highest predicted quality metric value.
17. The system of claim 16 , further comprising a transcoding module for transcoding the input image into the output image using the selected transcoding parameters corresponding to the highest predicted quality metric.
18. The system of claim 16 , wherein the array has been obtained from transcoding of a set of training images.
19. The system of claim 16 , wherein the set of feasible transcoding parameters comprises the resolution scaling factor zT not exceeding a maximum scaling factor z_max determined from predetermined maximum image dimensions of the output image and image dimensions of the input image.
20. The system of claim 16 , wherein the quality and file size prediction module is further configured to interpolate between entries of the array for at least one of the indices.
21. The system of claim 16 , wherein the quality and file size prediction module is further configured to limit the set of feasible transcoding parameters to those combinations of the output encoding quality factor QFT and the resolution scaling factor zT yielding predicted relative file size values not exceeding the maximum relative file size.
22. The system of claim 16 , wherein the array comprises a multi-dimensional relative file size prediction table indexed by the QF(I), QFT and zT.
23. The system of claim 16 , wherein the array comprises a multi-dimensional quality prediction table indexed by the QF(I), QFT and zT.
24. The system of claim 16 , wherein the input image and the output image are JPEG images.
25. The system of claim 18 , wherein the quality metric values are determined based on the Peak Signal to Noise Ratio (PSNR) measure of respective transcoded training images compared with corresponding training images in the set of training images.
26. The system of claim 18 , wherein:
the set of training images comprises a first set of training images and a second set of training images;
the predicted relative file size values are obtained from transcoding of the first set of training images; and
the predicted quality metric values are obtained from transcoding of the second set of training images.
27. The system of claim 18 , wherein the quality metric values are based on the Maximum Difference (MD) measure of respective transcoded training images compared with corresponding training images in the set of training images.
28. The system of claim 16 , wherein the quality-aware parameter selection module is further configured to sort the set of feasible transcoding parameters according to the corresponding predicted quality metric values, and truncate the set of feasible transcoding parameters by reducing the set to a promising subset corresponding to high predicted quality metric values.
29. The system of claim 21 , the quality and file size prediction module is further configured to limit the set of feasible transcoding parameters to those combinations of zT and QFT wherein the resolution scaling factor zT does not exceed a maximum scaling factor z_max determined from predetermined maximum image dimensions of the output image and image dimensions of the input image.
30. The system of claim 29 , wherein the maximum scaling factor z_max is further defined according to a viewing scaling factor zV chosen based on anticipated viewing conditions of the output image.