Thread: "16-bit" mode.
View Single Post
  #227  
Old November 29th 04, 07:37 AM
Dave Martindale
external usenet poster
 
Posts: n/a
Default

"Mike Russell" writes:

. In this 1995 article, Alvy states:
"Nonlinearity should never be stored in an image. Or, if it is, then this
nonlinearity must be noted in the storage format in such a way that it is
known how to remove it to retrieve linear data."


Right. All you need to satisfy the above is a way to decode the pixels
back to linear space. PNG has that. OpenEXR stores pixels in a version
of floating point - but the decoding method is specified. sRGB also
specifies how to convert between linear and encoded pixels. Looks like
this problem is mostly solved now.

Of course, you *can* leave data linear, but you need more bits for it.
The Pixar Image Computer, deveoped under Alvy at Pixar, used 12 bits per
colour component in memory, and in data files, while arithmetic was all
16/32 bit.

As for Timo's lonely stance - he appears to be in good company now, having
been debunked together with Alvy Ray Smith and Dan Margulis, all in the
space of a few days. :-)


Shall we then place George Preddy above all the others you mention? If
being debunked is a virtue...

Dave