Graphics Reference
In-Depth Information
advantage of offering more precision than the LogLuv formats.
2
The OpenEXR
format is very flexible. While RGBE and LogLuv are limited to three color chan-
nels, OpenEXR allows for arbitrary many channels. These extra channels can be
used for any number of things, such as compositing or general transparency. An-
other notable feature is the use of a 16-bit
half-precision
floating-point number
format. Ordinary floating-point numbers use either 32 bits (single precision) or
64 bits (double precision). The 16-bit half-precision format allows negative num-
bers, which may be needed by some color space representations. Furthermore,
some GPUs support half-precision floating-point computations so the OpenEXR
format meshes well with graphics hardware.
A drawback of OpenEXR is its complexity; however, ILM released a set of
software tools for working with the format along with the specification. The spec-
ification and tools were released under an open source license, which has encour-
aged the widespread use of OpenEXR outside ILM.
6.3 HDR Video
Since 2003, adapting HDR imaging to video has become more prevalent. This
section introduces some representative work in early HDR video capture that in-
spired a substantial body of work in recent years.
6.3.1 Video and HDR Images
The method of Debevec and Malik described in Section 6.1 requires a precise
knowledge of the exposure times used to capture the images in order to recon-
struct the HDR response function.
3
The assumption is that the aperture setting
of the lens was the same for all the images. Exposure control in video cameras
is more easily done by changing the aperture size, especially if changes need to
be made during filming. Image noise also tends to be higher in video shots, and
this disrupts the measurements of the pixel exposures across images. In the paper
“Radiometric Self Calibration,” Tomoo Mitsunaga and Shree K. Nayar presented
a new method for recovering the response curve of a camera using only rough
estimates of the camera parameters [Mitsunaga and Nayar 99].
2
For storing log
2
L
, the OpenEXR format has a higher precision than the 24-bit LogLuv, but lower
precision than the 32-bit LogLuv format. OpenEXR achieves a 0.1% quantization precision, whereas
32-bit LogLuv has 0.3% steps.
3
The authors were so concerned about accurate exposure times, they were not convinced the shut-
ter speeds provided by the camera were sufficiently precise. So they made audio recordings of the
shutter being released at each speed and then determined the actual speeds by analyzing sounds from
the recording.
