Photography and forensic science

Discussion in 'Digital SLR' started by Charles, Feb 7, 2010.

  1. Charles

    Charles Guest

    Ran across some interesting posts about faked photos. Led me to this:

    Then, ran across some opinions that RAW files cannot be faked? Wait a
    minute! The demosaicing algorithms are public knowledge. Algorithms are
    simply mathematical manipulations that surely are reversible. What am I
    missing here?
    Charles, Feb 7, 2010
    1. Advertisements

  2. Charles

    Paul Furman Guest

    Right, the raw files are a grid of red, green and blue pixels, when
    demosaiced, the green ones (twice as many) are given priority for
    lightness/darkness and the others are spread around to average things
    out. I don't know how you'd ever figure out how to un-average them back
    to the exact same places given each pixel has contributions from
    probably at least 9 adjacent pixels. You could make a raw file but not
    the same raw file. So you could make a totally fake picture but not
    modify a picture where authorities had a copy of the original raw.
    Paul Furman, Feb 7, 2010
    1. Advertisements

  3. Charles

    Ray Fischer Guest

    Ditto Canon. It's useful for police departments.
    Ray Fischer, Feb 8, 2010
  4. Charles

    Ofnuts Guest

    How does one detect that one is taking a picture of a printout of a
    doctored picture?
    Ofnuts, Feb 8, 2010
  5. Charles

    John A. Guest

    The thing is, when faking a RAW file you're not trying to reproduce
    the one-and-only original RAW. You're just trying to produce a RAW
    that *could* produce the JPEG/TIFF/whatever you have on hand.

    More than one possible RAW file to produce it? So what? Pick one that
    seems the most plausible and run with it.
    John A., Feb 8, 2010
  6. Charles

    Charles Guest

    I think they can and I think they will. My guess is that digital images,
    both RAW and JPG, will be weak evidence (or inadmissible) in many courts of
    law. Not an important issue to me ... just curious about it.

    Thanks to all who responded.
    Charles, Feb 10, 2010
  7. Then no email would ever be evidence, since it is so trivial to
    fake ... fingerprints are fakeable, too.

    Wolfgang Weisselberg, Feb 10, 2010
  8. Oh yes there is. Jpg images are 8-bit-per-colour, and while tif may
    have more bpc, if either are coded for viewing the levels defined by
    those bits will be evenly distributed in a display gamma space,
    typically around gamma 2.2 - 2.5.

    eg. in any image coded for viewing, levels 0, 45, 90, 135, 180, 225, 255
    will result in 7 almost evenly spaced shades of grey.

    Raw is encoded linearly, before any gamma for display is applied.
    Processing the raw image into jpg or tif applies a response curve, based
    on a gamma curve. This stretches out the codes in the shadows and
    compresses the codes in the highlights. Consequently, when all of the
    raw codes may have unique jpg and tif codes in the shadows many of the
    raw highlights map to the same code in the jpg and tif, especially with
    recent cameras with 14-bit ADCs.

    eg, with a 14-bit raw file, levels 2 & 3 would probably map to around
    levels 2 & 3 in a 8bpc jpg or tif, but level 15880 and next 45 or so
    levels in the raw would all map to level 252 in the jpg.

    Consequently, working back from the gamma space jpg or tif to the raw
    with linear space will leave hundreds of missing codes in the
    highlights, making it simple to differentiate true raw from one that has
    been reverse engineered from jpg or tif, even without manipulation. You
    can try to mask the missing codes using a variety of techniques, but
    they remain pretty easy to distinguish from real raw images.

    Then there is the characteristic banding and mapped hot pixels present
    in most raw images, which make it possible to prove that a reverse
    engineered raw could never have been created by the same camera as
    another real raw. Mapped hot pixels are pretty much undetectable to the
    end user, but are close to being a signature in every image that a
    camera takes.
    Kennedy McEwen, Feb 10, 2010
  9. Not quite.

    A chain of custody _is_ necessary if one needs to prove something about
    the physical photograph itself in a criminal trial. Such might be the
    case, say, if possession of the photograph itself is illegal or when
    forensic evidence is recovered from the physical print.

    It's rarely necessary when a photographic is being admitted for the
    image it purports to represent. In fact, usually the photographer
    need not testify.
    That can be done, and usually is done through witness testimony. It's
    often as simple as:

    Q: Witness, I present this object to you and ask you to examine it.
    Does this photograph fairly and accurately portray the scene you
    observed at the location and the time in question?
    A: Yes.
    Q: Your Honor, move to admit.

    In the case you cite, the court said "The photographer spoke to no one
    who was present the night of the incident, and no witness testified that
    the photograph was an accurate representation of the view from the
    Explorer into the Omni when the drug deal took place." The court
    did not even hint at a chain of custody requirement.

    If the photograph is the used to draw additional scientific results,
    such as the length of skidmarks, etc, the court can require additional
    technical information such as shooting data. In the case of an
    unattended photograph, testimony is normally required as to the
    proper operation of the equipment which produced it.

    Once admitted, of course, the opposing side is free to cross-examine
    and otherwise challenge the authenticity of a photograph. That then
    becomes an issue for the trier of fact to decide.
    Michael Benveniste, Feb 10, 2010
  10. Wrong guess. I have first hand experience that the digital images
    produced by speed cameras are exceedingly strong evidence in court! ;-)
    Kennedy McEwen, Feb 10, 2010
  11. Charles

    Charles Guest

    LOL. I should have said "some digital images ...." I live in Naples, FL
    and the new traffic cameras here are generating a nice income (for the
    company that installed them and for Collier county). They are mostly
    focusing (no pun) on right-turn on red violations (not coming to a
    full-stop) and that is really pissing off many drivers here. I have to
    agree as the many ugly crashes here are seldom due to right-turn on red

    Anyway, back on subject. Dedicated cameras for security and or violation
    detection are not what I was trying to get at.

    Thanks for your post.
    Charles, Feb 10, 2010
  12. Charles

    Charles Guest

    Well the list goes on and on, it would seem that there is no problem
    with photos from digital cameras being used in court.

    I predict many of them will be challenged in court. Who, what, when, where
    and why ... Even for shots from the hands of a forensenic CSI, the defense
    will be in a position to pose some chalenging questions. The fact that they
    "can" be faked is going to play a major role.

    Again, I'm not involved in any litigation and I am not a member of the legal
    system. I just find it to be an interesting issue.

    Have served jury duty several times and was amazed at how little details can
    become major issues and destroy an argument.
    Charles, Feb 11, 2010
  13. Charles

    John A. Guest

    And plausible values can be interpolated, semi-randomized (say with
    dark-frame data from the intended "source" camera), etc.
    John A., Feb 11, 2010
  14. Charles

    Paul Furman Guest

    Would it be possible to steal someone's jpg and create a fake raw file,
    assuming the original photog didn't shoot raw? You could fake the hot
    pixels for your own camera.
    Paul Furman, Feb 11, 2010
  15. Charles

    Joe Makowiec Guest

    Jens Duttke, the author of PhotoME did a proof-of-concept:

    He notes:

    Since this is only a proof of concept, an expert can still see,
    that the image wasn't taken using a real FZ-30. There are many
    indicators like noise, CAs, hot pixels, the limited color depth of
    JPEG etc. to determine that this image wasn't created by a digital

    But if I would investigate more time into this program, it would be
    easily possible to add noise, CAs and hot pixels to the image, so
    that even experts have problems to validate the image.
    Joe Makowiec, Feb 11, 2010
  16. Charles

    Peter Guest

    Of course, the cause of acquittal might also be due to prosecutorial errors
    and/or bumbling, think O.I. Simpson. It also might be good defense counsel,
    or even a belief in actual innocence. The press rarely presents exculpatory
    Peter, Feb 11, 2010
  17. Charles

    Peter Guest

    Overcoming CSI expectation adds an additional burden to the prosecutor.
    Failure to recognize this additional burden can lead to acquittal. But, it
    can be done and the prosecutor has an obligation to recognize this. Just as
    a defense attorney may very well play on the effect as an evidentiary
    Peter, Feb 11, 2010
  18. Charles

    Peter Guest

    Perhaps I was not clear. Needed or not, the lack of such evidence needs to
    be covered, just as both sides cover possible prejudice on the part of the
    Peter, Feb 11, 2010
  19. Not practical - faked missing codes don't match the noise homogeneity of
    a real raw, whatever means is used to create them. How would you know
    the hot pixels in your camera without disassembly?
    Kennedy McEwen, Feb 12, 2010
  20. Yes, if you convert the fake raw back to jpg then you get the same, or
    similar, continuum of levels in the jpg that you started from, but
    relatively trivial analysis of the fake raw shows that it is fake to
    begin with.

    I understood the original question related to the verification of the
    raw image. Shoot with jpg only and almost anyone can fake your images.
    Kennedy McEwen, Feb 12, 2010
    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.