Video Noise & Compression [OT](was Re: LRO: Land Rover video)

From: Jeff Berg (jeff@purpleshark.com)
Date: Fri Jun 29 2001 - 16:07:29 EDT

  • Next message: Ray Harder: "LRO: 95 rangie and on..."

    >I know what audio noise is, are you talking about that or video noise? If
    >VN, what exactly is it?

    I was talking about video noise and it's really the same thing as
    audio noise. The most basic definition of video noise could probably
    be stated as any visual artifact that isn't part of the "real" video
    information. White flashes--or dropout, degradation of the signal
    through generation loss, age/wear on the tape stock or problems
    introduced by faulty time base (hence the use of TBC's or time base
    correctors in professional edit suites) can all cause noise of one
    sort or another.

    The problem is that noise, being somewhat random by nature, is
    impossible to compress. All compression schemes are based on the
    notion that in any given series of video frames, many (most?) of the
    pixels don't change. A red land-rover driving across the frame
    against a "fixed" background is an example. So instead of storing all
    of the information for frame one, frame two... thru frame five
    million, four-hundred thousand and fifty-two the file just stores the
    information for the first frame and then stores the changes (or
    deltas) for the next series of frames. So our theoretical Red Land
    rover driving across the frame will compress better then a tracking
    shot that follows the red rover as it moves along the background.
    (Assuming that is that the red rover isn't dominating the frame...) I
    hope you get the idea.

    Every so often a complete or "key" frame is stored. Some compression
    schemes are very smart about when to do this. If the delta number
    exceeds a certain level, store the entire frame instead. An example
    of a new key frame might be a sudden cut from a dark shot of rovers
    at night to one of the group wheeling the next morning. As there is
    no relationship (from a visual perspective) between the two frames.
    Visually similar frames compress well, visually diverse frames do not.

    The problem with noise is that it introduces random patterns into the
    frame that disrupt the frame-to-frame continuity. This increases the
    number of deltas that need to be stored and may even cause each frame
    (in a particularly noisey sequence) to be stored in it's entirety.
    Result: Big files and "jerky" playback on slower machines.

    For the more technically minded, there is a much "cleaner" definition
    of compression in A. Michael Noll's book "Introduction to
    Telecommunication Electronics, Second Edition pp340-341, Artech House
    Publishers. (Disclaimer: I helped Dr. Noll with certain sections of
    the second edition of the book and am credited in the preface.)
    However, after reading it I think my "plain language" version might
    be more accessible--if more wordy--to the majority of the group.

    Sorry to those who got this far and are waiting for the Rover content...

    jeff

    -- 
    Jeffrey A. Berg     Purple Shark Media        Rowayton, CT        
                         jeff@purpleshark.com
                          ==================
    	I believe in heavy thinking, I believe in heavy sound,
    	I believe in heavy images, To hold it all down.
    				--Chris Smither, Happier Blue
    



    This archive was generated by hypermail 2b30 : Fri Jun 29 2001 - 16:37:26 EDT