Photography and forensic science

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

  1. Because by its very nature, the system forces such a bias on them.
    Rightly or wrongly, we judge the effectiveness of a police department
    by the "solve" rate. In the case of a crime lab, their funding,
    recognition and promotion depends on their work resulting in convictions,
    not on exoneration.

    In fact, the legal system recognizes that the bias exists and accounts
    for it procedurally. That's why the the rules for questioning under cross
    examination differs from that of direct testimony.

    And when it comes to investigating alleged internal misconduct, such as
    altering photos, police departments and other law enforcement agencies
    recognize the potential for bias as well. That's why internal affairs
    divisions exist. It's also why surveys consistently show huge majorities
    (80+%) of people saying serious complaints against the police should be
    investigated independently in addition to IAD.
    I find it ironic that I'm being accused of prejudice because I imply
    other people have bias. If you look at threads in different newsgroups,
    you'll find people accusing me of exactly the opposite bias. I've been
    challenged off of juries both by plaintiffs suing the State and defense

    Who's right? Which bias do I have? How am I supposed to tell?
    There's no way for anyone be objective about themselves. But from
    verifiable facts, you'll see that there's no real reason for me to have
    a bias either way.

    I've never been arrested, charged, or (as far as I know) investigated for
    any criminal activity. I haven't been cited for speeding ticket or any
    other moving violation for almost 20 years. I've been the victim of an
    armed robbery, of vandalism, and of credit card fraud, and in each case
    the police dealt with the situation professionally. I've been a witness
    for the prosecution in a felony theft case, and was treated fairly by all
    parties. My wife has been a consultant trainer for both the State Police
    and parole officers.
    That's the amazing thing about the adversarial system. Everyone thinks
    everyone else is biased. I firmly believe that everyone is right in that

    I will say this, though. Defense attorney's _love_ it when cops claim they
    aren't biased. Turning again to the O.J. case simply because most people
    know the players, Mark Fuhrman got caught by one of the oldest tricks in
    the book that way.

    We all have biases. While a professional tries not to let them impact
    their work, no one succeeds all of the time.

    But thanks for the civics lesson.
    Michael Benveniste, Feb 14, 2010
    1. Advertisements

  2. Charles

    Peter Guest

    I've been lurking in this discussion with interest. In real life I have seen
    both sides, actually three sides of the issue. Most in law enforcement are
    good people who try to do an honest job. there are some bad apples who
    succumb to temptations. (e.go. Many years ago the newspaper headlines
    declared that $350,000 was found at a crime scene. The defendant told me the
    amount was closer to $500,000. Investigative dropsy is not an uncommon
    occurrence in narcotics investigations. If a highway patrol officer does not
    issue his unofficial quota of tickets over an extended period of time, he
    will lose his seat. There are "pressures" to solve cases, especially
    sensitive ones. There are law enforcement people who have taken shortcuts
    that tends to exclude exculpatory evidence. Fortunately, this happens rarely
    and the work of defense counsel and IA investigators keeps the bad apples to
    a minimum. Just a short summary of my observations.
    Peter, Feb 14, 2010
    1. Advertisements

  3. The OP. In case you have forgotten, the question is recreating a
    believable raw from jpg.
    That gives you the dark *spatial* noise which, as explained several
    times, isn't the issue since the shadows are pretty much reproduced in
    jpg directly from the raw in the first place. Now, where do you get the
    noise for the mid tones and highlights? That is the noise that is
    needed to obscure those missing codes that appear when you go from
    something like gamma 2.5 space, required for the displayable image, to
    linear space. In a genuine raw it will be greater than the dark noise,
    from basic physics, but it will be much lower than the missing code gaps
    and will have a completely different spatial spectrum to the dark
    spatial noise.
    No it isn't - as explained, jpg compression results in very predictable
    spatial noise effects due to the DCT at the heart of the jpg coding
    algorithm. It is nothing like real noise in a raw image and does
    nothing at all to fill in the missing code gaps.
    Of course it matters. You seem to be assuming that it is perfectly
    filtered out - it isn't. How well each defective pixel is filtered
    changes from image to image due to the local scene content and
    illumination in the region of each defect. You can't extract the map
    from one image, even from 1000 images. However, with access to the
    defect map the identification of just one which is a single level
    different from that which should be computed by the concealment
    algorithm proves beyond doubt that the image is not a genuine raw from
    that camera.
    What would you propose to "smooth them out" with? Sampled sensor noise
    won't do it because the missing codes are much larger than the sensor
    noise amplitude. In the highlights, the difference between two adjacent
    codes in the source jpg results in 45 missing codes between levels in a
    curve corrected, linear 14-bit raw. The sensor noise for a decent
    camera is probably 5-10 levels. Add any more than that and your image
    is obviously fake.
    Point to one, then! I don't think there is any demand for it at all,
    hence this thread.
    Kennedy McEwen, Feb 14, 2010
  4. Charles

    John A. Guest

    Of course. And now we're discussing *how*. Specifically, we're
    discussing how to add realistic (as possible) sensor noise as would be
    expected in a RAW file.
    I believe you deleted the part of my post where I proposed getting
    noise source from a grey frame.

    If you think it will differ along the linear range from dark to light,
    feel free to take sample shots at several brightnesses and apply them,
    mapped by the de-noised source image.
    And so you remove that noise.
    Well it's good that there is a further test to authenticate a RAW
    beyond a cursory inspection, but how often are RAWs subjected to that
    level of scrutiny.

    I put forth that most claims of "I have the RAW image" would be
    checked pretty much by looking to see if it's visually the same image,
    and is in fact in the RAW format for some camera. And what we're
    talking about here goes a step or two beyond that. A perfect forgery?
    No. There is no such thing. Good enough to get away with it? For most
    cases, yes I think so.
    Increase the color-depth to 16 or 32 bit before removing the JPEG
    No demand for software to remove JPEG artifacts?
    (Add quotes around the search terms and there are still over 2000
    John A., Feb 15, 2010
  5. As already explained, so much data is lost in going from raw to jpg that
    it requires more noise to be added on the return that the noise level is
    no longer comparable with that of an original raw.
    It would be irrelevant what colour raw frame you recovered the noise
    from - it would be much lower amplitude than the noise you need to add
    to conceal the missing codes.
    Of course the noise varies with intensity, and more codes are missing in
    the highlights than in the shadows - the point is that little higher
    than the shadows you run out of enough noise to conceal the missing
    codes. So adding enough noise to do that makes the result clearly more
    noisy than a genuine raw.
    So you are left with missing codes - play on the swings or play on the
    roundabout, its all just play and no progress!
    Unlikely, since the whole point of raw is that it purports to be
    original data and cannot be viewed directly as an image.
    Anyone contesting the veracity of a raw image would be examining the raw
    data, its distribution in a full width histogram and the noise amplitude
    in that data.

    The OP wanted something that would fool all of the people all of the
    time. Your solution doesn't fool neither some of the people all of the
    time, nor all of the people some of the time but you think its enough to
    fool some of the people some of the time.

    That is a long way short what the OP asked for.
    Increasing the bit depth changes nothing, however suggesting that it
    does demonstrates how little you understand of the problem. Taking 8
    bit data and extending it to 16 or 32 bits is no different than adding
    decimal places to decimal data. You start with 1, 2, 3, 4 etc. and get
    1.000, 2.000, 3.000, 4.000 etc. instead. There is no more data between
    levels 3.000 and 4.000 than there is between levels 3 & 4, so when you
    convert to linear space you have *exactly* the same number of missing
    No demand for software to remove missing codes in reverse engineered
    Kennedy McEwen, Feb 15, 2010
  6. Charles

    John A. Guest

    Okay, you seem to be having trouble wrapping your head around this.

    More further on...
    But if most people think as you seem too, they may simply assume that
    synthesizing a RAW file is impossible, so why bother checking the one
    they are seeing for authenticity if it looks ok?

    I say that since there have been no RAW forgeries that any of us here
    have ever heard of, the bar is in fact that low.
    I think you're putting words in the OP's mouth.
    Try this as a proof-of-concept...

    1. Create a grayscale gradient from black to white at the highest
    color-depth available to you.

    2. Reduce the color depth, with dithering.

    3. Get a count of unique colors.

    4. Increase the color depth and count colors again - note they are the
    same, as you would expect.

    5. Apply any de-noising algorithm at your disposal to get a smooth

    6. Check the color count once again.

    I feel confident enough to let *you* check to see if this glove fits.

    You'll find that there are also colors added once you enlarge the
    image if the JPEG is at a lower resolution than the forged RAW. (In
    such a case you would produce a deliberately out-of-focus raw, and
    claim that that was why the JPEG was at a lower resolution.)
    So you're saying it can't be done because, to your knowledge, nobody's
    done it yet?

    I would say that the first attempts may be crude, but sufficient. As
    experience at detection and production increase, however, so will the
    quality. The methods I have suggested offer only a first
    approximation. They may never be close enough to perfect to evade any
    possible detection, but I think the bar will be raised before we come
    to the limits.
    John A., Feb 17, 2010
  7. Because I don't think that it is impossible to attempt to fake a raw
    file, just that it is impossible to succeed so it certainly is worth
    checking for authenticity.
    Fortunately you aren't a lawyer and I certainly wouldn't want someone as
    casual as you conducting my defence (or prosecution for that matter)!
    As Rumsfeld said "Absence of evidence is not evidence of absence!"
    The fact that both Canon and Nikon sell Original Data Verification kits
    strongly suggests that people have tried, despite what you have failed
    to hear.
    Dithering is not required for reducing the colour depth and adds nothing
    to this stage except add further noise.
    A smooth gradient at the expense, in this proof of "misconcept", of lost
    resolution. And I mean resolution, not pixel count! Same thing as
    happens with your jpg artefact removal.
    The colour count increases by as much as you are prepared to trade
    missing codes for resolution.

    Now try your same assessment with, instead of a simple greyscale, a
    pixel interleaved opposing greyscale. ie. all the even pixels
    transition smoothly from peak black on the left to peak white on the
    right and all the odd pixels transition from peak black on the right to
    peak black on the left. This has only one axis of full resolution, a
    real image has full resolution in both axes and in three colours.
    It would not be an "out of focus" raw.
    It would be a raw without focus.
    The two are very different - and obviously so.
    No, I am saying it can't be done because the laws of mathematics, not to
    mention the 2nd Law of Thermodynamics, forbid it. Something *you* seem
    to be having a problem "getting your head round".
    Kennedy McEwen, Feb 17, 2010
  8. Those odd pixels should, of course, be peak black on the right to peak
    white on the left. ie. in 8bpc, a 1024 pixel wide image from left to
    right would be:
    255, 0, 254, 1, 253, 2, 252, 3, 251, 4, 250, 5 .... 254, 1, 255, 0
    Kennedy McEwen, Feb 17, 2010
    Wolfgang Weisselberg, Feb 18, 2010
  10. You certainly do, consistently!
    Kennedy McEwen, Feb 19, 2010
  11. Interesting that you have NO WORD to add to the arguments
    against your arguments ... except some snide remark against
    my person. Does that mean I am right?

    Wolfgang Weisselberg, Feb 19, 2010
  12. No, you didn't post any arguments, just personal comment.

    Should I assume from your irrelevant response that you disagree that
    using a reduced data set jpg to create a raw which is indistinguishable
    from an (not *the*) original raw is a contravention of the second law of

    To do so would be to assume that you didn't have a clue, but I was being
    diplomatic and giving you the opportunity to rant in digression, as is
    your frequent want.
    Kennedy McEwen, Feb 19, 2010
  13. If you insist. So I posted a "just personal comment" on how
    your "X exists because people try to do something X is
    supposed to stop" arbument doesn't work in the real world
    based on observations in the real world.
    And what do you answer? Nothing.
    And what do you answer when poked? That:
    Please refer to my original posting before you ass you me.
    Your Honor, McEwen's ability to make up things other people
    supposedly said when they said something completely else is
    precious. I move he's to wear a red nose. Thank you.

    Wolfgang Weisselberg, Feb 22, 2010
  14. And that, as usual from you, is nothing but personal comment that was
    totally unrelated to the previous discussion.

    Your comment above has nothing whatsoever to do with people faking, or
    attempting to fake, raw files or the difference between the fake and a
    genuine raw - nor did your original post.

    Nothing I have said involves any compulsory process, as in your
    irrelevant comments. "X exists" in this case not because "people try to
    do something X is supposed to stop" but simply because the laws of
    mathematics *define* X's existence. It is entirely *optional* whether X
    is utilised - but I would expect any decent legal team to utilise the
    existence of X should it be required, rather than simply assuming that
    people don't do things that X can detect.

    The difference between a fake raw and a genuine raw exists - it isn't
    invented to stop people faking raw files! Any disputed raw file should
    be examined to determine if it is genuine and should not be assumed to
    be genuine simply because it is a raw file. Your comparison with air
    travel restrictions is irrelevant personal comment on a completely
    unrelated issue - I don't expect you to understand that, but I let you
    rant about it.
    You original post was irrelevant argumentative twaddle, merely
    demonstrating what an ass you are!
    You dispute your frequent rants on this forum? Googling the group's
    archive shows otherwise. I suggest you search on "Weisselberg" - you
    are quite unique and there is no shortage of irrelevant rants on your
    Kennedy McEwen, Feb 23, 2010
    Wolfgang Weisselberg, Feb 24, 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.