|
This comparative HDV
quality analysis between the 8-bit CineForm
Intermediate™ codec and the MPEG2 codec that
ships with Adobe Premiere Pro was performed at
CineForm's facilities by CineForm employees.
All components used are standard components
within Premiere Pro v1.5.1. This analysis uses
both visual inspection and image differencing to
determine visual performance characteristics.
The initial video
sequence came from the Sony FX1/Z1 demo reel.
The sequence of the red parrot was captured as
an MPEG2 stream and converted directly to
uncompressed RGB to become the master footage
for both work-flows. This uncompressed sequence
was placed on the Premiere timeline and exported
out as MPEG2 HDV (using the export-to-tape
option to generate an M2T file) and then out to
CineForm HD using Premiere Pro's file exporter.
Each of the resulting compressed outputs were
externally converted into uncompressed RGB AVI
files, ready for testing the next generation.
Using 8-bit RGB
intermediates in this analysis is a key
difference compared to the
quality analysis performed on our 10-bit
CineForm Intermediate which always
maintained files in a 10-bit uncompressed YUV
form. RGB is more common for prosumer editing,
and is required within Premiere Pro when using
MPEG2 files, so it is the proper testing
methodology for this type of equipment. As most
compression formats are YUV based (including
MPEG2 and CineForm HD) the conversion from YUV
to RGB (and back again) introduces a small loss
that occurs in each generation. These losses
are presented consistently in both workflows.
Titles were added
using the internal Premiere titling tool. Each
generation used the uncompressed RGB result from
the previous generation. The steps for testing
the MPEG2 workflow:
-
place the RGB
master in the timeline.
-
add title “1” .
-
export to M2T
(MPEG2 generation 1).
-
convert M2T to
RGB as MPEGgen1.avi.
-
place
MPEGgen1.avi on a new timeline.
-
remove the first
frame (i.e., start timeline on the second
frame of MPEGgen1.avi).
-
add title “2” .
-
export to M2T
(generation 2).
-
convert M2T to
RGB as MPEGgen2.avi.
The steps between 2
and 6 are repeated for each generation.
CineForm Intermediate testing follows an
identical procedure. In addition to CineForm
Intermediate and MPEG2 testing, a parallel
workflow that exported directly to uncompressed
AVI files was used for the differencing tests.
The step involving
removal of the first frame through each
generation (step 6) is designed so that motion
compression used in both MPEG2 and CineForm HD
have to recalculate a new ‘I’ frame fore each
generation. Without this operation, both
compressors wouldn’t have to work very hard –
and compression artifacts would accumulate
slower. This simulates the natural editing
process which makes editing on ‘I’ boundaries
only a 1 in 15 chance in 60i MPEG2. For those
who are wondering, CineForm HD does use a form
of motion compression (this is why CineForm
files and smaller than other ‘I’ only
compressors) so its compression is also effected
by this operation. CineForm HD uses a two-frame
GOP (group of pictures) for enhanced editing
performance; it has a 1 in 2 chance of aligning
with the “GOP” boundary for any random edit.
This one frame slip test slightly favors MPEG2
over CineForm HD, as the CineForm workflow never
gets to exploit its short-GOP advantage (GOP
alignment occurs far more commonly in CineForm
HD editing, resulting in even better
multi-generation performance.) An improved
analysis could use a random slip size, just like
real editing.
The difference
images were calculated by comparing the
uncompressed workflow to that of the two
compressed workflows. The difference images
where created by extracting uncompressed frames
from the timeline (in RGB.) These frames were
loaded into an image-processing tool (GIMP)
where the source and resulting compressed images
were "differenced" (not subtracted which can
hide errors.) These difference images were
lifted to a middle grey by setting the black
level to 128 using a curve control, a luminance
region where the eye is more sensitive. These
color images were de-saturated so the magnitude
of the errors is clearly shown. Unlike our
10-bit CineForm Intermediate quality analysis
there is no further amplification of the errors,
as MPEG2 errors are clearly visible without
additional gain. |
