Re: `Lossless' retroactive embedding of color profile (avoid JPEG recompression); basic color profile theory
Re: `Lossless' retroactive embedding of color profile (avoid JPEG recompression); basic color profile theory
- Subject: Re: `Lossless' retroactive embedding of color profile (avoid JPEG recompression); basic color profile theory
- From: Graeme Gill <email@hidden>
- Date: Wed, 02 Aug 2006 17:33:30 +1000
Marco Ugolini wrote:
Even the highest quality setting in JPEG causes a detectable amount of
damage to a file, which is compounded with each successive action of saving.
This isn't strictly true with most encoders and decoders. The point is, that
once the image has been quantized (in DCT/YCbCr space), it will be decoded
and encoded to the same quantized values. (A decoder that adds random dithering
or similar on decode, or if the quality level is changed each time etc., will
continue to degrade the image with each successive cycle).
Now because the quantization is in a transformed space of RGB, the rounding
errors involved can nudge values from one quantized bin to the next a bit,
but after a few rounds these changes will probably settle down.
It's easy enough to do the experiment (and I've done it). Get an original,
save it as JPEG in Photoshop with a given compression. Close it. Open it.
Save it as TIFF. Close it. Open it. Save it as JPEG etc. Use the Image/Calculation
to diff the plane values from successive TIFF's. Track the number of pixels
in the diff that are not 0. Each time, fewer and fewer pixels change.
While this amount of damage at the highest setting is not serious for most
purposes (and less than it used to be a few years ago), it may become
noticeable after being compounded by a number of saves.
I think you're absolutely right, if you're making any sort of global
adjustment each time. If you shift the pixel values and then
re-compress them, you add another bunch of initial quantizing error.
Graeme Gill.
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Colorsync-users mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
This email sent to email@hidden