NetNews Usenet Archive 1993 #3/NN_1993_3.iso/spool/rec/photo/24281

Text File | 1993-01-24 | 2.9 KB | 60 lines

Newsgroups: rec.photo
Path: sparky!uunet!zaphod.mps.ohio-state.edu!sdd.hp.com!hpscit.sc.hp.com!hplextra!cello!jacobson
From: [e-mail] (David Jacobson)
Subject: Re: How Autofocus works?
Message-ID: <[e-mail]>
Date: Sat, 23 Jan 1993 17:07:06 GMT
References: <[e-mail]>
Organization: Hewlett-Packard Laboratories
Lines: 49

In article <[e-mail]> [e-mail] (Marcell Stoer) writes:
>I would like to exactly how the autofocus works in the new SLRs.

The AF system consists of two linear CCD sensors, some optics, and a
processor.  Usually the main mirror is half silvered or perforated so
some of the light passes through it, bounces off a secondary mirror,
and then falls on the CCD arrays, which are the same optical distance
behind the lens as the film is.  Some additional goodies (note the
precision of my description :-) cause one of the sensors to see only
light passing through the left side of the lens, whike the other sees
only light passing through the right side of the lens.  If the image
is properly focused, the two ccd arrays see the same thing.  But if
the lens, is, say, too close to the film, the image picked up by the
left CCD array will be shifted a little to the left relative to the
one in the right CCD array.

Ostensibly the algorithm is a correlation (as a function of shift)
algorithm.  Here it is in a nutshell.  Subtract the average light
intensity from the intensity at each point.  This gives positive and
negative numbers.  Let us call the ones from the left sensor Li, for i
running from 1 to 128 or however many elements the CCD array has, and
the right sensor values Ri.  Now denote the shift by s.  Compute the
function F(s) = Sum over the range of i of L(i-s)*R(i+s).  Obviously
there are a lot of little technical problmes to be solved, for example
what do when when i-s or i+s is off the end.  But anyway, when it is
perfectly focused Ri = Si for all i, and for s=0 we are either multiplying
positives times positives or negatives times negatives and we get a
a nice large postive value for the sum.  But if the image is out of
focus the maximum will occur for some non-zero values s.  Assuming
that subscripts increase to the right, if the max occurs for a
positive value of s, the lens must be moved farther from the film, and
vice versa.

Now, again, let me emphasize that there are lots of little technical
problems whose solutions are no doubt closely held secrets.

>Does anyone know exactly how this works?  If there are some publications
>on this I wouldn't mind the references.

If you have easy access to patents, try finding the original patent,
which is held by Honeywell.

>Marcell Stoer                                Internet : [e-mail]
>Department of Chemistry
>Molecular Beam Laser Spectrometry Laboratory    Voice : [phone]
>University of Victoria                            FAX : [phone]

  -- David Jacobson