iafilm.co.nz Home
This article is licensed under a
Creative Commons Attribution 4.0 International License.
You may quote and use on condition
that you attribute John Calder as the source of this info.
What's up with Film
Digitisation?
by John Calder
In the 1980s, John Calder was a
champion of New Zealand
Indie Super-8, making 4 x Super-8 features and a legion of shorts.
His later "indie" movies are shot on "Hi-8" and
"Mini-DV". Day jobs have included commercial film production
management, teaching, computer programming and truck driving. His day job
now is teaching computer multimedia and programming at "MITCIT.NZ". (Manukau Institute of Technology). He is currently
experimenting with (a) animation using digital stills cameras, (b) digitising
his 1980's legacy
and (c) doing High Definition movies by shooting on film then digitising.
This article is about (b) and (c).
The Issues
- Digitising the legacy - there
must be a quality loss on copying, but lets have
some actual measurements on how much loss with different methods and what
we need to do to get a high quality result - I will aim for the copy
retaining 90% or more of the line resolution of the original.
- "Indie" Film-making
now - Standard Definition Digital Video (Mini-DV) is very accessible but
not quite good enough for what I want to do. I want to do
"greenscreen" effects shots and DV's low resolution colorspace
tends to plant green edges around our foreground objects. I head to
the cinema to see "indie" movies shot in DV and react with
"almost good enough resolution but not quite". The
introduction of lower-cost High Definition cameras raises a
challenge. Pour money into an early generation of new technology?
Or shoot with dear old film and digitise that? And which film -
Super-8 or 16mm?
Why
build my own digitiser?
I want
creative control, sometimes down to customising individual shots and even
individual frames. Also I have an idea on how I can do a high quality job
= automate the taking of an individual photograph of each frame with a digital
stills camera. And I want to make this kind of digitisation
affordable. There are film industry processes based on individual
photographs of frames, but they are very expensive.
Main points of the machine.
�
Built around a
gate+claw from a dual gauge Super-8 + Standard-8 projector.
�
Runs very slowly on a
geared down motor. Takes 6 secs between frames because that is how long
the digital stills camera takes to take and store a shot.
�
Straight-through path
with no sprockets is kind to damaged archival film.
�
Uses mid-range Olympus
Camedia 750 camera. Uses a lens from a 35mm stills camera as a
close-up lens.
�
Automatic run achieved
by a Microswitch triggered by the gate motor which fires an adapted infrared
remote.
Test
film.
I shot
Super-8mm black and white film of a standard EIA1956 line resolution
chart. The film stock, Kodak Plus-X Reversal asa40 is one I could process
myself to get an immediate result.
How good
is Super-8mm ?
Looking at
the result under a microscope gives an EIA1956 result of 600 line pairs both
horizontally and vertically. This compares well with the best of SD video
at 540 horizontally and about 450 vertically.
With this
indication of Super-8 being about 10 per cent better than SD video (720 pixels
wide) then I will suggest that Super-8 is roughly equivalent to a digital
imaging medium of 800 x 600 pixels.
Fig
1: My EIA1956 result from Super-8mm.
Fig
2: FOR COMPARISON - EIA1956 result from a Sony DCR-TRV950 3-chip DV video
camera. This is generally regarded as a result of "540".
Quality of a digitised copy.
The "guts" of this investigation. What happens when we digitise
film at various copy resolutions? I first tried to checked
this out by using the Olympus zoom lens to vary my photos of the Super-8 image.
Fig
3: First Test method (representative 4 images)
BUT the
zoom-out results were lower than previous tests I had done with video
cameras. On double-checking by taking full-frame shots while changing
resolution with the camera controls, I got higher resolution results.
This opens
another interesting question which I will ask by example.
To get a
final result of 720 x 576 do we shoot an image of that size first-off?..
OR is there any benefit to shooting a higher-resolution image first off then
downsizing later in the computer? I come to this expecting
that it should make no difference, but this first bad result makes me re-think!
I therefore
took my final set of measurements full-frame at varous resolutions and used a
little zoom-out - approx equivalent to the first image above - to get some
in-between settings. Note that the EIA1956 resolution reading is a rather
subjective art. Figs 1 and 2 show that well.
The stills
camera, an Olympus Camedia UZ750, offers 2 quality settings for jpg compression
"High" and "Normal". Trying both, I found
they made no difference for shooting at 1280 x 960 and above. But below
1280 the "High" showed up as delivering an advantage. The
graph shows the "High" settings with one "rogue" point for 720 x 576
coming from a different setup using a video camera.
Results
ImageID
|
Copy
Image Width
|
Copy
JPG
Setting
|
RES
|
File
Size (KB)
|
Notes
|
2
|
2288
|
HI
|
600
|
2723
|
|
3
|
2048
|
HI
|
600
|
1534
|
|
4
|
1600
|
HI
|
600
|
1058
|
|
5
|
1280
|
HI
|
600
|
754
|
|
B5
|
1280
|
NORM
|
600
|
272
|
Economy
file size works well here
|
6
|
1024
|
HI
|
570
|
498
|
|
7
|
1024
|
NORM
|
550
|
164
|
Economy
quality loss effect appears
|
10
|
844
|
HI
|
550
|
394
|
|
VidCam
|
720
|
DV
|
430
|
160
|
|
8
|
640
|
HI
|
520
|
184
|
|
9
|
540
|
HI
|
430
|
181
|
|
I see
enough of a pattern here to suggest a "rule of thumb".
IF there is a difference between the Resolution of the Original
and the Resolution of the Digitising Method
THEN - the Copy inherits the lower Resolution of the 2
ie "A chain is as strong as
its weakest link"!
ELSE - IF the Original and the Digitising method are
similar in Resolution
THEN - the Copy is of reduced Resolution, by between 10
and 30 per cent - but we need to investigate more.
|
The results
are better, and this rule is more optimistic, than what I was expecting from my
past studies of maths and physics formulas. Although
this may fit the "Nyquist-Shannon Sampling Theorem" (Wikipedia, 2006,
1) if anyone could come up with a meaningful definition of the
"bandwidth" of Super-8.
The
greatest uncertainty happens in the "ELSE" zone, which is where one
of my biggest questions lies = what reasonable final result can we expect from
copying Super-8 to Standard Definition TV? (720 x 576).
My answer
is to come up with a reasonable workflow to produce the common DV flavour of
SDTV then look at the resolution.
For copying
on my machine with the Olympus Still Camera, I see that the best "bang for
my buck", ie smallest file size giving a 600 res result, comes at 1280 x
960 with the "NORMAL" (ie low) jpg setting. Files are about 270KB each which
means I would usually be able to fit 1 x 50foot Super-8 roll on 1 x 1 Gig
memory card. This has a lot of practical appeal because it means
transferring those common rolls in one unattended automatic run without needing
a stop for a change of card.
So to get
720 x 576 PAL, or the NTSC 720 x 480, the
workflow that appeals is to shoot at 1280 x 720 then process to 720 x 576. Lets do it and see what happens�
The
resolution is fairly good at about 470. You can see how tricky it gets
judging that point where we lose the distinctiveness of the lines, and it seems
to be especially hard in this "ELSE" area of approx matching
resolutions. I could argue for results from anywhere in the range 430 to
500.
But Ouch! the "aliasing" effect is a bit of a worry!
Look at the staircasing on those diagonal lines! I need to rethink my
workflow idea.
I applied 3
video processing filters to my upside-down negative original.:
"Flip - Vertical", "Invert" (turning a negative to a
positive image), and "Brightness/Contrast" (it is normal for raw
negatives to be low contrast and I need to increase the contrast in further
processing)
Retain
"Flip" only to keep it readable, and the result is STILL BAD ..
"Flip"
etc are provided by a Video Editing Program, "ULead MediaStudio Pro
version 7". This is an old version of ULead, so to be fair I try
again, downloading the trial version of their current Version 8. VERSION
8 WORKS OK but I know my boss won't buy it because video editing is only a
small part of our work and his budget is tight this year.
I look for
an open source or low cost answer.
Next Experiment:
Try a different "Flip" - this is the one provided by "The GIMP
version 2.2.10". Looks like "The GIMP" does a good
"Flip".
And here is the makings of a workflow. Fortunately "The
GIMP" is scriptable in that it can consistently "Flip"
collections of images on demand.
Lets try
again. New workflow.
1.
Work in "The GIMP" at 1280 x 660 to process the image(s)
2.
Then load results into the Video Editor to downrez and process to SDTV.
Result:
Form your
own opinion, but I am going to declare at 470
First interesting
"indie" conclusion:
Prediction
of what happens to film copied to SDTV
Super-8mm
470
16mm
540 (lesser of the 2 media delivers its maximum)
Super-16
540
35mm+
540
Armed with
this result, I now try to improve my math model graph.
I replace
the (720,430) point with this new (720, 470) result then overlay some lines to
represent the rule-of-thumb, with major weighting given to the hard-earned
(720, 470).
The Cusp
(point of max correction factor) happens at (800,500)
The
Original and Copying-Method resolution are both 600. The expected
resolution is 500 which is 83%. This is the
maximum correction factor.
I now try
to predict other film format calculations by re-scaling the Cusp to
(100, 83) so I can apply this as as percentages of any film digitising
scenario�
To predict
the behaviour of 16mm, I will first need to find its line resolution.
Practical
Experimentation with "Simulated 16mm"
I know that
16mm is bigger and better than Super-8mm. But I want to know HOW MUCH
better with this digitisation workflow?
Shooting
16mm tests with fresh film stock was out of the question for this investigation
because of the cost. The minimum quantity I could buy and process was 10
mins costing NZ$350 to do the test.
What I did
instead was to shoot tests with a 35mm Stills camera, aiming to frame so that
the charts and other test shots occupied a space of about 10mm x 7mm.
I then cut
16mm wide strips out of the processed negatives and placed them in a projector
gate to photograph.
I used
consumer Fujicolor film rated at asa 200.
My frame
sizing did not hit the mark. The chart read about 560 but appeared about
2/3 the size of the projector gate it was sited in. Applying
a correction for that SUGGESTs an EIA1956 resolution rating of 830 for
Standard-16mm at 200asa.
So that is
600 - for
Super-8mm
830 - for
Standard-16mm (but with a more flexible film stock)
I will now
do a wild guess for the effect of a finer-grain film and suggest
the
following EIA1956 resolution rating summary.
600 -
for Super-8mm
900 -
for Standard-16mm at 100asa
1080 -
for Super-16mm at 100asa
What is
this in Pixels? Now getting very approximate, and it will vary a lot with
different film stocks, but I am going to regard 16mm as follows:
Standard-16mm
- 1200 x 900
Super-16mm
- 1500 x 900
This
suggests that even Standard-16mm can hold its head high as a genuine High
Definition Digital Television format.
But what is
going to happen to 16mm on digital copying and HDTV projection?
I now look
to the generalised graph for answers.
Listed
below are my predictions for the various fates of Super-8mm, Standard-16mm and
Super-16mm. Units are EIA1956 line resolution. See Appendix One for
the math detail.
HDTV-720 is
the 1280 x 720 medium.
HDTV-1080 is the full 1920 x 1080 medium just emerging as of early 2006.
I have not included HDV which is an intermediate 1440 x 1080 anamorphic medium
rescaled to 1920 x 1080 for viewing.
By
"HDTV" I mean a large TV or video projector wired to the playback
machine
ie "Closed Circuit" as in a film festival scenario, without quality
losses due to broadcast.
Appendix One gives the details of how I use the graph to get these
values.
Super-8mm (without cropping, appears with
black side bars on 16:9)
Original
600
SDTV
470
HDTV-720
540
HDTV-1080
600
Standard-16mm (without cropping, appears with
black side bars on 16:9)
Original
900
SDTV
540
HDTV-720
650
HDTV-1080
810
Standard-16mm - cropped to 16:9 for HDTV.
The horizontal rez calculations give higher values but the loss in vertical
resolution probably results in a human perception of about the same resolution
as the 4:3 display.
Original
900
SDTV
540
HDTV-720
760 (650)
HDTV-1080
900 (810)
Super-16mm - cropped slightly to 16:9
Original
1080
SDTV
540
HDTV-720
840
HDTV-1080
990
Based on this I now draw some conclusions to
guide me and my friends in our own film-making. 16mm conclusions first
and I wrap up
with my beloved Super-8mm
(It is up
to you how much weight you give to my test methods and indie point of
view!)
Standard-16mm makes good sense from where I stand
as the entry path into High Definition Digital Moviemaking. The cost of
film and processing competes well with buying or hiring HD Video Cameras for
the next 10-minute movie I have lined up. Within my "indie"
subculture there are good Standard-16mm cameras available. My likely crew
of fellow 50-somethings dragged kicking and screaming out of suburbia to relive
their hippie alternative cinema youth, are well skilled in 16mm. Like all
film options, the digitisation workflow is going to make post production far
nicer than we had it in the 1980s.
Super-16mm. I do not have access to
this. One friend has toyed with the idea of converting a camera, probably
at a cost of NZ$1000. My advice to him will be "don't bother, spend
the money on film stock and get on with it, Standard-16mm rocks for what we
want to do". To make 16mm go in my "indie" subculture,
part of the make-it-affordable equation will be DIY digitisation via home-made
machinery, and this will be much easier for Standard-16mm. I can
temporarily alter the light source in a borrowed 16mm projector and run it with
a strapped-on car windscreen-wiper motor, but hacking at the gate with a file
would be a no-no!
For those
working with a higher level, Super-16mm makes a lot of sense. Wikipedia
states under "16 mm Film":
"Some low-budget theatrical features are shot on Super-16; ironically, thanks
to advances in digital technology - specifically digital intermediate (DI) -
the format now seems to be seen as revitalized option." - (Wikipedia,
2006, 2)
Should I
make new movies with Super-8mm?
Should You make new movies with Super-8mm?
The Super-8
frame is 5.3mm wide
The
Standard-16mm frame is about 10mm wide
The
Super-16mm frame is about 12mm wide.
My results
suggest that Standard-16mm, although twice as big as Super-8mm, has only about
1.5 times the resolution, mainly because of the common use for 16mm of more
sensitive film stocks of lower resolution. I suggest that
Super-8mm "punches above its weight" but as a cheeky Super-8
film-maker from way back I would enjoy stating that. Some more evidence
would be nice.
On Sept
17th 1985 I produced Colin Rock's script "Do It With
a Real Fox" in 16mm but also with some Super-8. I have found closely
matching shots of performer Linc Mail from the 2 cameras and put them through
my digitising process. Here they are! Printed equal frame height -
cropped narrower to fit side by side.
Guessing
game - which is Super-8mm and which is 16mm? The Answer is revealed at
the end of this article!
I will
almost certainly make at least 1 new Super-8mm short film to explore the
option. But it will be with subject matter, an offbeat short fantasy
drama, where the artistic quality of grainy texture will work for the story.
I'll be doing this mostly for the nostalgia revisit. I want the
experience of working with a Super-8 where many of its classical handicaps are
gone but the fun remains.
I've kept
the cameras and the processing tank all these years, gotta do something with
them. With Kodak's new Super-8 Ektachrome 64T film to try out, full
computer post-production power, and effective projection commonly available at
film festivals, that could be a very interesting something.
Should
YOU make new movies with Super-8mm?
My message
to most young emerging film-makers would be "no". Resist the
"retro romance" reputation given to Super-8mm by some avant-garde
artists and take your own good hard look at it. Use it only if you want
that particular "look" or if the most easily available and most-liked
camera for you is a Super-8. When I do Super-8mm, I know that in terms of
conventional ideas of image quality, I will be
below the edge that 16mm can give me at about half the price. If you are
paying for transfer to video, the price gap will narrow. 16mm will then
give genuine high definition at about only about 30 percent greater cost than Super-8mm.
My Super-8
rez results are higher than what I expected, but come in the range of
"almost good enough to compete, but not quite".
Another
warning to the romantics- beware the gate-registration
curse of Super-8mm.
The easy-load
cartridge has film transport disadvantages which can show up as vertical
jiggling of the image. This varies a lot with cameras and
conditions. During these tests, my steadiest results came from a Canon
814E.
But note
that 16mm may mean in practice using cameras over 50 years old, and that also
means a risk of jiggle ("poor gate registration").
Test well
if you can. I have!
Follow-up - June 2007
Since writing this article I have produced a new Super-8 film,
"Love Song to Super-8",
partly to try out Kodak's new Ektachrome 64 film stock.
Some new information: I found this colour stock gave me very grainy results. It suited this particular
movie but it also delivered less resolution than the black and white film used for the tests
reported on above. My updated opinion is that Super-8 can deliver high definition results but ONLY
with fine grain black and white film eg Kodak 7276, Kodak 7265. Otherwise my conclusions and warnings remain the same as above.
Even the mighty 16mm is increasingly under threat from affordable Hi-Def Video with
the new
Canon HV20 Camera looking like an excellent breakthrough tool for no-budget
indies.
How best
to archive Super-8?
I have my
own answer. Digital Still Photos at 1280 x 960 pixels.
There has
been vigorous debate on whether owners of real-time condenser lens systems
should upgrade their video cameras to high definition. eg
Sony FX1, Canon XL-HD1.
"I
have about 4 hours of super 8 video, flimed in the seventies and eighties. Some
of the movies we brought to a "professional" transfer to DVD service
but the results were not very good.
My wife
was just about to bring them to another shop, but then I said, let's use the
HDR-FX1 and find out about the quality of the transfer. No telecine device was
used. Just an old Elmo Super 8 projector onto a 2 feet
diagonal screen. FX1 was on HD, filming directly off the screen.
What did
we get? Stunning results! I was afraid first about strobing effects due to the 24
fps projector speed and the 60i recording. But there was practically no
strobing. So far the best transfer of Super 8 movies I have ever seen."
Internet
Discussion: Author alias "hptschupp", (sonyhdvinfo.com, 2005)
"the resolution of Super 8 is too low in my opinion to
benefit from HD"
Response to
above: Author alias "cprogrammer", (sonyhdvinfo.com, 2005)
I suggest
that it is indeed worthwhile to use HDV cameras like the FX1.
Super-8 original capability
|
600
|
Copy with SDTV camera - "480i" or
"576i"
|
430 - 470
|
Predicted result with HDV camera - "1080i"
|
570
|
Appendix One:
The summary table in full. More details on where those numbers come
from.
Super-8mm (without cropping, appears with
black side bars on 16:9)
Original
600 - from direct
experiment as above
SDTV
470 - from direct
experiment as above
HDTV-720
540 - 960 pixels � 120% � 90% x 600
HDTV-1080
600 - 1440 pixels
is in the 1280-and-over range
Standard-16mm (without cropping, appears with
black side bars on 16:9)
Original
900 - experiment
as above (830) with estimated
correction for finer grained film.
SDTV
540 - expect full
SDTV performance
HDTV-720
650 - 960 pixels � 80% � 72% x 900
HDTV-1080
810 - 1440 pixels � 120% � 90% x 900
Standard-16mm - cropped to 16:9 and therefore
using full horizontal width
(but it loses some vertical resolution, note these values are horizontal res
only)
Original
900
SDTV
540
HDTV-720
760 - 1280 pixels � 106% � 84% x 900
HDTV-1080
900 - 1920 pixels � 160% � 100% x 900
Super-16mm - cropped slightly to 16:9
Original
1080
SDTV
540
HDTV-720
840 - 1280 pixels � 85% � 78% x 1080
HDTV-1080
990 - 1920 pixels � 128% � 92% x 1080
Reference
List
Wikipedia, 2006, 1. "Nyquist-Shannon Sampling Theorem",
Wikipedia On-Line Encylopaedia Website. Retrieved 30
Apr 2006
from
http://en.wikipedia.org/wiki/Nyquist-Shannon_sampling_theorem
Wikipedia, 2006, 2 "16 mm Film", Wikipedia On-line
Encylopaedia Website. , Retrieved 30
Apr 2006
from
http://en.wikipedia.org/wiki/16mm_film
SonyHDVInfo.com, 2005, "Super-8 to Video Project",
SonyHDVInfo Discussion Groups. Retrieved 30 Apr 2006 from
http://www.sonyhdvinfo.com/showthread.php?t=3588&highlight=Super-8mm
Guessing
Game Answer
Super-8 on the LEFT, 16mm on the RIGHT.
Super-8 is
Agfa Moviechrome asa40, camera is Sanyo XL600
16mm is
Kodak Ektachrome Video News Film asa125, camera is Bolex H16R
iafilm.co.nz Home