| | |
BitJazz SheerVideo Test Data Measurements
In this table, we show a thumbnail of each numbered image
in the order the images appeared on the original Kodak PhotoCD.
Below each image, we show
These figures are given for the most popular uncompressed formats:
all of which SheerVideo directly supports.
Which, by the way, no other perfect-fidelity codec does.
thumbnail:
In this table, we show 96x64 GIF thumbnails of the numbered images to give an impression of their level of detail.
The thumbnails were created by reducing the original images using Adobe Photoshop's bicubic interpolation
and converting to indexed color mode using a selective palette of 256 colors.
This is about the best GIF can do.
The thumbnails aren't in Sheer format because we can't expect you to install SheerVideo before learning about it.
We chose GIF because, after the initial image degradation from converting an image to indexed color,
GIF compresses and decompresses it with perfect fidelity.
JPEG, in contrast, degrades both the color and the spatial detail,
and browsers vary widely in their JPEG decompression fidelity.
PNG, as a perfect-fidelity RGB codec, would be superior to GIF,
but older browsers unfortunately don't support it.
uncompressed size:
The raw uncompressed size given here consists of just the picture,
and excludes all header and other packaging information.
Moreover this uncompressed size consists of just the visible pixels,
excluding all line or frame padding such as horizontal and vertical blanking intervals.
Furthermore, the uncompressed size quote here consists of just the intrinsic components,
and excludes all component padding and pixel padding.
For example, each of the original TIFF images contains a 1620-byte header
which the uncompressed sizes listed here strip off.
And on the Macintosh, uncompressed QuickTime movies
in RGB and Y'CBCR 8bv 4:4:4 formats
always contain an extra unused byte per pixel,
as if they were RGBA and Y'CBCRA 8bv 4:4:4:4,
but this empty alpha channel is ignored in the uncompressed sizes quoted here.
Thus the uncompressed size as defined here represents just the bare components of the bare pixels
of the bare scanlines of the bare frames, and is calculated as
uncompressed size = component precision x pixel depth x frame width x frame height / 8 bits/byte
SheerVideo size:
To get the frame sizes of the Sheer-encoded frames,
we dropped the compressed movies on Dumpster and opened up the sample-size ('stsz') tables,
which give the exact number of bytes in each frame of the QuickTime movies.
This is also a quick way to verify the frame size for the uncompressed movies.
Note that this measure excludes all non-image information,
except that it rounds up the total compressed size to the nearest byte.
If you want to include all the QuickTime padding,
multiply the size of each uncompressed RGB or Y'CBCR 8bv 4:4:4 frame by 4/3
to account for the unused alpha channel.
To include all the frame-specific non-image QuickTime header information
in the size and compression calculations,
add 4 bytes (for the frame offset) to the size of each uncompressed frame,
and add 8 bytes (4 for the frame size and 4 for the frame offset) to the size of each Sheer-encoded frame.
Dumpster is a handy little free utility from Apple
that lets you look inside the guts of a QuickTime movie file.
You can download Dumpster
for Mac or Windows from Apple's developer website, http://developer.apple.com/.
But unless you really know what you're doing,
it's a good idea to lock your movie files before opening them with Dumpster,
to keep from accidentally corrupting them on account of Dumpster's aging and quirky user interface.
In particular, Dumpster doesn't properly filter keyboard shortcuts for commands and doesn't support undo at all,
so if you click in an editable text field and type 'Command-S' to save a change,
Dumpster will replace the contents of the field with the letter s.
And if you then type 'Command-Z' to try to recover the original contents of the field,
Dumpster will mindlessly add the letter 'z' after the 's'.
power:
The compression power, also known as the compression ratio, is straightforwardly defined as
power = uncompressed size / encoded size
Note that a compression power of 1.0 means no compression at all.
Some people like to put an 'X' or a ':1' after the number, as in "2.0X" (two-times power)
or "2:1" (two-to-one power).
space savings:
The amount of space you save by using SheerVideo instead of uncompressed video,
as a percentage of the uncompressed size, is defined as
space savings = (1 - encoded size / uncompressed size) x 100%
error:
The root-mean-squared error (RMS error) of the compression+decompression cycle is defined as
the square root of the sum of squared differences between the reconstructed image and the original image,
where the sum is taken over every sample of every pixel of every frame:
rms error = (Sumsample(reconstructedsample - originalsample)2)1/2
That may sound complicated, but the RMS error is actually the simplest uniform measure of error.
Literally, it just measures the Euclidian distance between the original and reconstructed images.
To express this in bits, simply take the base-two logarithm:
error = log2(rms error)
SheerVideo is a perfect-fidelity codec,
so it always has zero error.
In other words, there is never any discrepancy at all between the original and reconstructed pixels.
In fact, SheerVideo even has an optional Self-Check mode
that you can turn on to verify each compressed frame
by decompressing it and comparing the reconstructed image with the original.
This could be useful if you have flaky hardware,
or if your computer is exposed to unusual radiation.
|
|
|