JPEG detectors and JPEG image history estimators
An image is classified respective to JPEG compression history. An image signature is derived from the image. The image signature is indicative of spatial frequencies content of chrominance components of the image including spatial frequencies content associated with JPEG compression. The image is classified respective to whether the image has undergone JPEG compression based on analysis of the image signature.
1. A method for classifying an uncompressed image respective to JPEG compression history, the method comprising:
deriving an image signature from the uncompressed image, the image signature being indicative of spatial frequencies content of one or more chrominance components of the uncompressed image including spatial frequencies content associated with previous JPEG compression; and
classifying the image respective to whether the uncompressed image has previously undergone JPEG compression based on analysis of the image signature;
wherein the method is performed by an apparatus for classifying an image respective to JPEG compression history.
2. The method as set forth in claim 1 , wherein the deriving comprises:
suppressing spatial frequencies content of the one or more chrominance components that is associated with luminance gradients.
3. The method as set forth in claim 2 , wherein the suppressing comprises:
smoothing the one or more chrominance components along high values of a luminance gradient.
4. The method as set forth in claim 2 , wherein the suppressing comprises:
weighting the one or more chrominance components based on a corresponding luminance gradient.
5. The method as set forth in claim 1 , wherein the deriving an image signature comprises:
generating horizontal and vertical projections of at least one of the chrominance components; and
transforming the horizontal and vertical projections into a frequency space.
6. The method as set forth in claim 5 , wherein the deriving an image signature further comprises:
combining the transformed horizontal and vertical projections to produce a combined projection.
7. The method as set forth in claim 5 , wherein the generating of horizontal and vertical projections and the transforming thereof is performed for two chrominance components, and the transforming further comprises:
combining one or more of the transformed horizontal and vertical projections of the two chrominance components to produce a combined projection.
8. The method as set forth in claim 1 , further comprising:
for each uncompressed image determined to have previously undergone JPEG compression, estimating an image re-sizing subsequent to the previous JPEG compression based on frequency values of JPEG compression artifact features in the image signature.
9. The method as set forth in claim 1 , wherein the classifying comprises:
processing the image signature with a cascaded classifier, the processing terminating when either a classifier stage of the cascaded classifier determines that the uncompressed image has previously undergone JPEG compression or no classifier stage determines that the uncompressed image has previously undergone JPEG compression.
10. The method as set forth in claim 9 , further comprising:
tagging each uncompressed image determined to have previously undergone JPEG compression with a JPEG quality based on which cascaded classifier stage made the determination.
11. The method as set forth in claim 9 , further comprising:
training each classifier stage of the cascaded classifier using a training set including uncompressed images that have not previously undergone JPEG compression and uncompressed images that have previously undergone JPEG compression.
12. The method as set forth in claim 11 , wherein at least two classifier stages of the cascaded classifier employ different classification algorithms.
13. The method as set forth in claim 11 , wherein each successive stage of the cascaded classifier is trained using a sub-set of uncompressed images of the training set that are not classified as having previously undergone JPEG compression by earlier trained classifier stages of the cascaded classifier.
14. An apparatus for classifying an image respective to whether the image has undergone JPEG compression, the apparatus comprising:
an image signature deriver that derives a spatial frequency-based image signature from at least one or more chrominance components of the image; and
a classifier that classifies the image respective to whether the image has undergone JPEG compression based on analysis of the spatial frequency-based image signature.
15. The apparatus as set forth in claim 14 , wherein the image signature deriver derives the image signature with suppression of at least some spatial frequencies content not associated with JPEG blocking artifacts.
16. The apparatus as set forth in claim 15 , wherein the image signature deriver comprises:
a fast Fourier transform (FFT) that transforms one or more image components to a spatial frequency space.
17. The apparatus as set forth in claim 16 , wherein the one or more FFT-transformed image components are selected from a group consisting of a luminance component, a first chrominance component, and a second chrominance component.
18. The apparatus as set forth in claim 14 , wherein the classifier comprises:
a plurality of connected classifier stages.
19. The apparatus as set forth in claim 18 , wherein the plurality of classifier stages are cascaded, and each successive classifier stage is more likely to classify the image as having JPEG compression than the previous classifier stage.
20. The apparatus as set forth in claim 18 , further comprising:
a tagger that tags the image with a JPEG quality indication corresponding to a classifier stage that classifies the image as having undergone JPEG compression.
21. An image classification method comprising: deriving a spatial frequency-based image signature from one or more chrominance components of an image that is not JPEG compressed, the image signature comprising a spectral representation of the digital image; and classifying the image respective to whether the image has previously undergone JPEG compression based on the spatial frequency-based image signature comprising a spectral representation of the digital image;
wherein the image classification method is performed by an image classification apparatus.
22. The image classification method as set forth in claim 21 , wherein the deriving further comprises:
adjusting the one or more chrominance components based on a corresponding luminance component to suppress chrominance component features attributable to luminance component gradients.
23. The image classification method as set forth in claim 21 , wherein the deriving comprises:
computing a spatial frequencies representation of one or more projections of components of the image.
24. The image classification method as set forth in claim 21 , wherein the classifying comprises:
training a classifier using a training set of image signatures derived from a set of uncompressed images and images that have been JPEG compressed; and
applying the classifier to the derived spatial frequency-based image signature to classify the image respective to whether JPEG compression has been performed.
25. The image classification method as set forth in claim 21 , further comprising:
generating a plurality of rotated images from the image, the deriving and classifying being repeated for each of the plurality of rotated images; and
classifying the unrotated image as having previously undergone JPEG compression if at least one of the plurality of rotated images is classified as having previously undergone JPEG compression.