Graphics Reference
In-Depth Information
0.6
G
reen
0.5
Red
0.4
0.3
v
CCIR 709 Color Space
0.2
Blue
0.1
Visible color space
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
u
Figure 6.7
A slice of the CIE Luv color space (
L
=
1). The inside triangle is the color gamut of the
CCIR 709 color space, which is used in Kodak's PhotoCD format. (After images courtesy
of Greg Ward.)
tends to look like a series of uniform steps in brightness. In contrast, human color
response is closer to being linear, so
u
and
v
are unaltered in the format. For
his new format, Ward decided to use a variant of the Luv color space, in which
the
L
value is replaced by its base-two logarithm. Storing the logarithm of
L
in the
pixel representation accomplishes essentially the same thing as the “exponent” in
the RGBE format, and it better matches human perceptual response to brightness.
Tone mapping was another reason he chose to use the logarithm of
L
. Tone map-
ping typically works with the luminance rather than the color channels; in fact,
some operators use the logarithm of the luminance directly. Using the Luv space
in the file format not only facilitates tone mapping computation, it mitigates pos-
sible precision loss that might be incurred in color space conversion. Ward's new
HDR file format, known as LogLuv, has found widespread use, and is now part
of the TIFF specification. The format has two variants.
LogLuv format (32 bit).
An Luv pixel is packed into 32 bits by using 16 bits
to store log
2
L
, and 8 bits each to store
u
and
v
. The values are all represented
as fixed-point integers. While
u
and
v
are never negative, log
2
L
needs a sign bit.
Figure 6.8 illustrates how the bits are allocated. The first bit is the sign of log
2
L
,
the next 15 bit are its absolute value, the next 8 bits store
u
, and the last 8 bits
store
v
.The“
e
” subscript indicates the fixed-point representation. This format
provides a better representation than the 32 bit RGBE format, even though it uses
the same amount of storage per pixel.
