home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: sci.image.processing
- Path: sparky!uunet!haven.umd.edu!ames!saimiri.primate.wisc.edu!zaphod.mps.ohio-state.edu!news.acns.nwu.edu!casbah.acns.nwu.edu!lapin
- From: lapin@casbah.acns.nwu.edu (Gregory Lapin)
- Subject: Re: Pixel averaging paradox (help!)
- Message-ID: <1993Jan22.162425.13738@news.acns.nwu.edu>
- Followup-To: sci.image.processing
- Sender: Greg Lapin
- Nntp-Posting-Host: unseen1.acns.nwu.edu
- Organization: Northwestern University, Evanston Illinois.
- References: <1j781hINNih7@shelley.u.washington.edu>
- Date: Fri, 22 Jan 1993 16:24:25 GMT
- Lines: 32
-
- In article <1j781hINNih7@shelley.u.washington.edu> dean2@tbone.biol.scarolina.edu (Dean Pentcheff) writes:
- >Image analysis people: I've got a puzzling problem.
- >
- >Here's the setup: For various obscure reasons, I'm doing some analysis
- >on some videotapes. Using a realtime video analysis system (Datacube),
- >I'm trying to get average pixel intensities in a set of nested areas.
- >The areas are 1x1 pixel, 2x2, 4x4, 8x8, and 16x16 pixels. The 1x1 area
- >is in the center, the 2x2 framed around it, the 4x4 framed around that,
- >etc.
- >
- >On each video frame I'm reading the (B&W) pixel intensities from all
- >the pixels in each of the boxes. For each box, the pixel intensities
- >are summed and divided (all floating point - not integer) by the number
- >of pixels in the box.
- >
- >Here's the puzzling problem. During "boring" times (when there's
- >little action in the video image) the average intensities are: 2x2 >
- >4x4 > 1x1 > 8x8 > 16x16. Huh? Why should this be? Based on several
- >tries on various images, this is consistent. The same pattern is
- >somewhat evident when images are moving around, but of course there are
- >confusing effects due to bright patches hitting parts of the outer
- >boxes, but not some of smaller, inner boxes, etc.
-
- How about calculating standard deviations along with your means. You will
- probably find that there is no statistically signficant difference between
- the measurements.
-
- Just a guess, but you may be seeing a noise difference between video
- fields.
-
- >-Dean
- Greg
-