[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: DCT (forwarded)

At 5:57 PM 4/30/95, Rob Lingelbach wrote:
>--- Forwarded mail from "Richard Torpey" <rich at pi.mte.com>
>From: "Richard Torpey" <rich at pi.mte.com>
        No argument - the DCT is fast as a bat outta H*** in the edit suite,
>shuttles and locks up faster than anything but a ddr. However, for a graphi=
>suite 99 times out of a hundred it would be fine - that one 1 a hundred job
>would give me ulcers though - Do I feel lucky today, huh, do I?
>      =20

Rich, I have to tell you that we have yet to see any difference, or=
 *measure* any difference between input and output on our DCTs, whether used=
 for telecine, graphics or whatever.=20
        The principle relies upon the amount of correlation in a picture:=
 every picture that Ampex came across, that was not a specially designed=
 "break-it" test pattern, (i.e. any picture which conveys meaningful picture=
 content), has greater than about 2.5:1 correlation in the transform domain=
 where the compression acually occurs.=20
        This means that only *truly* redundant information is thrown away,=
 leaving picture material untouched. That's not virtually untouched, but=
 completely untouched. The reason it works is that we are dealing with=
 real-world pictures. Even computer programs can be compressed *without=
 loss* by a ratio of over 3:1. The "Stuffit" program on my Mac regularly=
 achieves this ratio, and yet I can still run the decompressed applications=
 fine. Try running "Stuffit" on a "Pict" file and see the compression ratio=
 that can be achieved without trying.
        The result is that we dont have to feel "lucky". We can (and do)=
 guarantee total transparency with all *legal* D-1 program material. The=
 only downside to the whole process is that we cannot allow any errors to=
 creep in at all. One error will corrupt an entire 8x8 block for a field,=
 and be embarrasingly visible. Ampex, to their credit, have turned a vice=
 into a virtue: they simply had to include an incredibly powerful=
 error-correction mechanism in the DCT specification, to get around this=
 potential mega-error problem. As it happens, this makes DCT actually=
 *cleaner* than most D-1s which are error-concealed, not corrected.
        There, I told you Mike Arbuthnot had done a number on me, and=
 really, that Swiss bank account number he slipped into my pocket helped a=
 great deal too  :-)

P.S. anyone out there care to dispute the above? Havent had a good argument=
 in ages.........