<html><head><meta http-equiv="Content-Type" content="text/html charset=us-ascii"></head><body style="word-wrap: break-word; -webkit-nbsp-mode: space; -webkit-line-break: after-white-space;">On Thu, Feb 13, 2014 at 8:08 PM, Joseph Ashwood <<a href="mailto:ashwood@msn.com">ashwood@msn.com</a>> wrote:<br><br><blockquote type="cite"><br>From: Phillip Hallam-Baker</blockquote><blockquote type="cite">I have a solution to the random number generator problem that can be<br><blockquote type="cite">built for about $50 and is completely verifiable.<br>[shake a flask of candy, take a picture]<br><br></blockquote><br>I'm not confident it will have as much entropy as you think. The design is<br>a fairly basic modification of the lavarand design.<br><br></blockquote><br><div>Put aside the question of camera digitization noise and just consider the random placement of discrete objects in the camera's scene. For simplicity replace the flask with a flat bottom box, such as a shoe box, and maybe rectangular candies, small compared to the box. Standard machine vision algorithms can measure the edges, and hence position and orientation of such objects to sub-pixel precision, easily one part per thousand in X, Y, and Theta with a megapixel camera. We don't have to make the measurements, just know the information is in the image. That's 30 bits per object, assuming the objects are all visually distinct. (If the objects are identical visually, subtract log2(10!), or about 22 bits.) And then there is the position and orientation of the camera, which can also be calculated from the image. So a photo of ten different objects, randomly placed should easily produce at least 256-bits of entropy before even considering other information in the image. </div><div><br></div><div>Arnold Reinhold</div></body></html>