Image Forensic Analysis

Discussion in 'Open Discussion' started by RobWatson, Jun 12, 2013.

  1. RobWatson

    RobWatson Mu-43 Hall of Famer

    Not really, but lately I've been getting increasingly interested in the topic.

    For fun I 'created' a "Attacking Dragon" image upon which to apply analysis "prove" the fakeness.

    Raw + jpeg here

    Attacking Dragon
    <a href="">
    <div align=center>
    [​IMG]"600" height="800"

    Hackerfactor is a nifty site.

    There is lots of info on the web and professional options as well.
  2. RobWatson

    RobWatson Mu-43 Hall of Famer

    Interesting paper. Not sure I agree with everything contained.

    In general for shapes unknown it can be very difficult. Start with a known shape (like a ball) and work towards a generalization. For example a ball out of focus will have a blur generally radially symmetric with no curl - see gradient operator and note that the divergence is radially symmetric. In general if the motion is directly towards/away from the camera (particularly on the optical axis) then focus and motion blur become impossible to differentiate - maybe.

    Presumably if one can clearly identify focus blur for a ball then generalize to other geometric shapes a linear mixing or even principle component analysis methods could be used to extend the ball results to objects on unknown shapes. Residuals would be assigned to be motion blur. Seems like something worth fleshing out in more detail ...