method to reveal photoshop manipulations

Discussion in 'Digital Photography' started by sobriquet, Sep 25, 2007.

  1. sobriquet

    sobriquet Guest

    Suppose I have a picture and I modify the picture in photoshop to
    remove an element in one part of the picture and add an element in
    another part of the picture. Subsequently I save the original with the
    same settings as the modified picture and I remove the meta-info
    (exif, etc..) of both pictures with a utility (e.g. jhead).
    Is there any method to more or less objectively determine which of
    these two pictures is the original and which one has been manipulated?

    Below I have two pictures for instance, one of which has been
    manipulated. Can you guess/determine which one of them has been
    modified?

    http://www.ibbu.nl/~nsprakel/x.jpg
    http://www.ibbu.nl/~nsprakel/y.jpg
     
    sobriquet, Sep 25, 2007
    #1
    1. Advertisements

  2. sobriquet

    Eric Miller Guest

    My guess:

    It appears that x.jpg has been modified by adding the cyclist and other
    parts of the scenery from y.jpg. This modification makes it appear that the
    cyclist has not changed position between the two images which were taken at
    different times.

    Anyway, I think that in answer to your question, that the answer is probably
    negative. I suspect that image manipulation is largely a subjective,
    judgment call. But the cavernous abyss that is my lack of knowledge about
    software and how changes to images may be analyzed by it should be taken
    into consideration along with my opinion.

    I do think that there is an objective means to determine that certain images
    have not been manipulated, but that requires certain equipment manufactured
    for forensic photography and doesn't really answer your question.

    Eric Miller
    www.dyesscreek.com
     
    Eric Miller, Sep 25, 2007
    #2
    1. Advertisements

  3. sobriquet

    pisnahuj Guest

    pisnahuj, Sep 25, 2007
    #3
  4. sobriquet

    just bob Guest

    My guess is both. No EXIF in either file so it's probably not the original
    camera file.
     
    just bob, Sep 25, 2007
    #4
  5. sobriquet

    Ali Guest

    Done well, but here's what I think:

    The cyclist in the brown coat doesn't look right.
    The shadow on the white van doesn't look right.
    The road position of the van in the closer shot looks a bit strange.

    So 'Y' (the one with the van furthest away) is the closest one to the
    original shot.
     
    Ali, Sep 25, 2007
    #5
  6. sobriquet

    Pat Guest

    I don't have a lot of time to look, but it appears that Y is the
    original.

    On the X file, there's a thin white line under the van that appears to
    be a crop line.

    Also, under extremely low contrast, the white of the van matches the
    other whites but on the X file, the van is whiter than the rest of
    your whites.

    the true tests are performed by science journals that can to
    algorithms on them and test the files. also, a close zoom-in can
    detect the differences in pixel patterns around the van in one
    location or the other.
     
    Pat, Sep 25, 2007
    #6
  7. sobriquet

    sobriquet Guest

    The cyclist is part of both pictures, so it must be part of the
    original.
    The white bus has been edited out from one position and has been
    edited in at an adjacent position. The question is whether the white
    bus was on the left or right of the cyclist in the original picture.
    You can closely examine the difference between the original and the
    manipulated image by opening them in photoshop (or gimp) in a single
    document with each of the pictures on a separate layer and zoom to
    100% or more visibility while turning the top layer off and on or
    changing the blending mode.
    http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=389x1761730

    (same as http://tinyurl.com/25tp6r )

    But this method doesn't seem to work in this case. When I save the
    manipulated image at a slightly lower jpg quality, subtract that from
    the manipulated image, apply levels to make the differences between
    the images visible, it doesn't reveal the parts of the image that have
    been manipulated.
    It probably only works if the images used to manipulate an image are
    sufficiently different and not photo's that have been taken under
    virtually identical circumstances.
     
    sobriquet, Sep 25, 2007
    #7
  8. sobriquet

    sobriquet Guest

    Yes, but that is not the kind of manipulations I mean. Both images
    were saved as jpg with identical settings and both had their exif and
    other meta info stripped.
    But only one of them was saved without any modifications in photoshop
    (the original jpg from the camera was saved under another name without
    doing anything to it apart from saving it, while the other has been
    manipulated to blend in elements from another picture taken under
    virtually identical circumstances).
     
    sobriquet, Sep 25, 2007
    #8
  9. sobriquet

    sobriquet Guest

    But it was part of the original picture and has only been modified on
    one side where it blends in with the bus.
    On both sides of the cyclist?
    Either way, the van that has been blended in had its shadow copied
    from a virtually identical photo, so it looks as it should look,
    although subtle variations in the jpg encoding or other subtle
    differences between the pics or how the elements have been blended in
    can make it stand out more potentially.
    Both road positions must be kind of 'factual' though, since it was not
    put in manually at an arbitrary spot, but the spot was determined by a
    photo taken under virtually identical circumstances where the van
    happened to be in that position, while the bicycle and all other
    moving elements where in other positions.
    Indeed.
    Here is the original shot (y.jpg):

    http://www.ibbu.nl/~nsprakel/img_2380.jpg

    And the pic that has been used to modify it:

    http://www.ibbu.nl/~nsprakel/img_2381.jpg
     
    sobriquet, Sep 25, 2007
    #9
  10. sobriquet

    Ali Guest

    His right leg looks soft and his left leg has a black outline. Where did
    the black outline come from if the light source is from that side? Maybe
    the scale is a bit off too, but it sometimes difficult to judge in a 2D
    image.

    The shadow is very narrow. Also, the light source must be very low. What
    is the light source? And, why aren't there any other similar shadows.

    Initially I was originally going to say that the camera was on a tripod and
    there were two shots taken (hence the change of position of the van). But,
    I didn't post this because of the shadow and van position. FWIW, I
    originally thought the van may have been turning left (combined with the
    fact that the driver and passenger were looking left). But, I'm not so
    sure, it just doesn't look right.



    Sometimes you can't quite put your finger on it, but know what is an
    original image.
     
    Ali, Sep 25, 2007
    #10
  11. sobriquet

    Dave Cohen Guest

    I really don't care which one has been manipulated. Just an observation
    though, PS is one of many available photo editors any of which are
    capable of the same manipulation, however, none of them will set you
    back $600 or so. Just a thought.
    Dave Cohen
     
    Dave Cohen, Sep 25, 2007
    #11
  12. sobriquet

    sobriquet Guest

    But wouldn't it be interesting to have a method to detect whether or
    not an image has been manipulated in photoshop?
    Suppose someone anonymously mails you a picture of a significant
    other in a compromising situation, it might be useful to have to
    ability to figure out if the picture has been manipulated or not.
    PS has some advantages though:

    1) it's free (you can download it from p2p)
    2) you can find many great tutorial videos on PS on p2p
    3) as far as I know photoshop offers more advanced ways to
    manipulate images compared to an alternative like gimp

    (points 1 and 2 might be illegal depending on local copyright laws)

    Or are there perhaps gimp fans out there who can confirm you can edit
    layer masks as custom rubylith overlays in gimp which is one of the
    things I love about photoshop.
    One vexing issue with PS (CS3) though is that while doing this, I
    can't turn on/off the visibility of the entire layer (while turning on/
    off the layermask is possible) without switching away from the
    rubylith overlay mode.
    It can also be slightly buggy in some situations:

    http://groups.google.nl/group/adobe...r:[email protected]+rubylith&rnum=1#897700ce1188ce81

    (same as http://tinyurl.com/29gqbu )
     
    sobriquet, Sep 25, 2007
    #12
  13. sobriquet

    Eric Miller Guest


    Not true. Because the photos were taken from the same position, the
    cyclist could have been added to either photo.
    Perhaps, but the "white bus" is from two separate images. This is
    apparent by the fact that the occupants of the white bus are in
    different positions in both images.

    Yes, I will do that when I have time.

    Eric Miller
    www.dyesscreek.com
     
    Eric Miller, Sep 25, 2007
    #13
  14. sobriquet

    Good Man Guest

    From 2004: http://www.hypography.com/article.cfm?id=34254

    Basically, an algorithm is used:

    "Farid's algorithm looks for the evidence inevitably left behind after
    image tinkering. Statistical clues lurk in all digital images, and the ones
    that have been tampered with contain altered statistics.

    "Natural digital photographs aren't random," he says. "In the same way that
    placing a monkey in front of a typewriter is unlikely to produce a play by
    Shakespeare, a random set of pixels thrown on a page is unlikely to yield a
    natural image. It means that there are underlying statistics and
    regularities in naturally occurring images."

    Farid and his students have built a statistical model that captures the
    mathematical regularities inherent in natural images. Because these
    statistics fundamentally change when images are altered, the model can be
    used to detect digital tampering."
     
    Good Man, Sep 25, 2007
    #14
  15. sobriquet

    sobriquet Guest

    Logic dictates that if one of the two pictures is the original, and
    the (same) cyclist is part of both images, it dictates that the
    cyclist must have been part of the original image. Only the elements
    that change between the images as you alternate between the original
    and the modified version are potentially not part of the original
    image.
    One of the white vans was part of the original picture and the other
    has been copied and pasted from a nearly identical picture (where the
    van and it's occupants were in a different position).
    But as far as I can see there is no infallible argument that implies
    which of the two vans has been edited out or was part of the original
    picture. For instance, some people might falsely assume that since the
    van is approaching the photographer, the picture of the van that is
    nearest must be the one that has been edited in.
     
    sobriquet, Sep 25, 2007
    #15
  16. sobriquet

    just bob Guest

    If the exif has been stripped I'm going to assume it's a fake. That's
    basically how the Canon verification software works.
     
    just bob, Sep 25, 2007
    #16
  17. sobriquet

    sobriquet Guest

    But a photo does not become fake by just having its exif info stripped
    or modified, when the actual visual content of the photo remains
    unaltered.
     
    sobriquet, Sep 25, 2007
    #17
  18. There's one telltale that X has been altered less than perfectly: there
    is a too-dark edge between the cyclist's jacket and the white van
    behind it. The limited resolving power of the optics and the
    anti-aliasing filter in the camera mean that the pixels located on the
    boundary between jacket and whatever is behind it will contain a
    mixture of light from the two sources.

    If the van had really been there, and you magnified the edge, you'd see
    the darkness of the pixels in the coat smoothly get brighter and become
    the bright pixels of the van. Measuring intensity along a line
    perpendicular to the edge and graphing it should give an S-shaped curve.
    But instead the pixels on the boundary are darker than either the coat
    or the van (because they are really a blend of the coat with the dark
    car in the distance). This dark edge should not be there.

    But it's possible to avoid this artifact with sufficiently clever
    software. Essentially, the edge needs to be processed in a way that
    figures out how much foreground and background is present in each pixel
    (to determine the alpha value for the mask that will be used) and then
    subtracts out the background leaving the foreground appearing as it
    would against a black background. Then the foreground can be composited
    over a new background with the edge appearing correct.

    Dave
     
    Dave Martindale, Sep 26, 2007
    #18
  19. sobriquet

    Ron Hunter Guest

    PS is NOT free, but, like just about anything, including inground
    swimming pools, CAN be stolen.
     
    Ron Hunter, Sep 26, 2007
    #19
  20. sobriquet

    JL Guest

    The x has been manipulated.

    With both photos, it is more easy to say which was manipulated.
    If only x is show, and if we don't want which element is suspect, it is very
    difficult to say if it was manipulated because they are too numerous
    elements to verify.

    Cordialement,
    Jean-Luc Ernst
    www.hopitaldelaphoto.com
     
    JL, Sep 26, 2007
    #20
    1. Advertisements

Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.