Müller HM 73

[Will2]

I've been seeing more and more about this machine on the web. Does anyone own one here?

Regular 8, Super 8, 9.5, 16 and Super 16mm all on the same machine.

Looks like about a $30,000 investment with the wet gate but after some avisynth scripting it seems to deliver amazing results.

[joelpierre]

http://www.repaire.net/forums/film-arge ... uper8.html

03/05/2011, 17 h 54
Test 2 : Die Filmverstatt (scanner HM73)

http://www.repaire.net/forums/film-arge ... 1970021320

http://www.repaire.net/forums/film-arge ... 1970021358

[carllooper]

The Muller demo reel is exceptional. Particularly love the Regular 8 from 1928. Gorgeous.

http://www.youtube.com/watch?v=PMlJXyfRmYk

[David M. Leugers]

Regular 8mm from 1928? R-8mm didn't arrive until 1932. The B+W clip looks so good, suspect it might be 16mm which was around then (1923 intro).

[joelpierre]

The Muller demo reel is exceptional.

It is a "demo" commercial. A "real" test is less bluffing :

http://www.repaire.net/forums/film-arge ... 1970021320

[carllooper]

The Muller demo reel is exceptional.

It is a "demo" commercial. A "real" test is less bluffing :

http://www.repaire.net/forums/film-arge ... 1970021320

Great test. Of course the Muller is a film transport system on the one hand, and an interchangeable camera on the other. Change the camera and you need to do another test.

[aj]

mr Müller constructs a website under the pretence of consumer information to recommend and promote his fantastic machine. http://www.smalfilmtest.nl/ When looking up DNS of the site it shows that the site is owned by Müller

All machines besides the Müller get very negative childish remarks and are called fraud or dishonest etcetc. Kunee (professional who even had a booth at the Photokina) is blamed for not sharing info. Is that an obligation to competion? Moviestuff becomes toyish qualification. MWA poor image quality poor over all. Fumeo perf ripper ???
Of course he actually didn't use all these machines for testing and it is merely hearsay or convenient invention.

The müller thing is a reverse-engineered MWA with some choices of its own which is then presented as an invention of himself.

[bolextech]

André, that is exactly the impression I got when I saw the smalfilmtest website.

The repaire.net evaluation does give a more unbiased evaluation.

According to the repaire tests, the MWA and the Müller do share one characteristic: registration is not great. They make it up somewhat in post with all kinds of stabilization software.

On the positive side, these types of continuous running scanners may be the best avenue for transferring severely damaged films.

My guess is that if the footage is in good shape, a system with conventional sprocket and claw would most probably give better registration.

Cheers,
Jean-Louis

[carllooper]

According to the repaire tests, the MWA and the Müller do share one characteristic: registration is not great. They make it up somewhat in post with all kinds of stabilization software.

Mechanical registration isn't very accurate on any machine, be it camera, projector, or scanner.

My feeling is that a scan should not, in the first instance, try to use (whether mechanically or digitally) any registration method at all. It should just try to reproduce the film as is, ie. as a long strip of film.

The way to do this is treat each scan as reproducing some arbitrary area of the film relative to another arbitrary area of film, rather than relative to the sprocket hole (or frameline, or content). So long as each scan overlaps another it is then just a matter of digitally stitching the entire film together to a form a single virtual strip of film, in the same way panoramas are stitched together from individual images.

The beauty here is that the film is reproduced as is, whether it has damaged sprocket holes or not, whether it has floating framelines or not. You have avoided the need for any registration at all.

Once you have a virtual film strip you can then implement decisions about how the virtual film is to be registered (for projection/viewing), be it through sprocket recognition, frameline recognition and/or content recognition.

Carl

[bolextech]

Hi Carl,

That's an interesting way of looking at it.

I guess it would be somewhat like the stabilized Zapruder film:
http://www.youtube.com/watch?v=TKcUL1EtbCI

What is the Müller not doing right then if there are complaints of instability?

Cheers,
Jean-Louis

[carllooper]

What is the Müller not doing right then if there are complaints of instability?

I wasn't meaning to suggest that the Muller implements what I was talking about. The Muller registers the film by relying on the mechanical registration method used by the camera: sprocket holes. The idea I'm proposing is not to use any registration method at all. At least not for the scan.

Any "instability" in the Muller must be a result of either a faulty camera, torn sprocket holes, or the laser not otherwise recognising the sprocket correctly. The real solution is not to introduce any dependance on sprocket holes (mech reg) at all.

I guess it would be somewhat like the stabilized Zapruder film:
http://www.youtube.com/watch?v=TKcUL1EtbCI

What I was really talking about is how you scan the film in the first place: assembling a copy of the film that doesn't rely on any registration at all. The stabilised Zapruder film is registered. It is registered with respect to the content, which is something you can always do after scanning the film.

If you could visualise the scan data I'm proposing, it would simply look like this (ignoring the watermark):



How you then project/view that scan data is a secondary step, independant of the first. You could use the captured sprocket holes, the framelines, or as the Zapruder film does: some registration point in the content.

The main idea is to avoid baking any registration method into the scan. Once the scan is done (which would be the same process for all films) you can then employ a problem specific registration method - whatever the film might require.

Technically the scan/stitch method should probably be called "self registration".

[Nicholas Kovats]

Carl,

According to their web site they in fact do not utilize a sprocket based system but a combination of PTR roller tension and laser alignment via the film perfs, i.e.

http://www.filmfabriek.nl/technology.html

"Drive & PTR

The machine works without sprockets and grabbers so that even shrunken and difficult to scan films can be digitized. The film is maintained on tension with rollers. The scanner comes with 4 PTR rolls to remove dust and hairs effectively from the film"

Laser alignment for stabilized image

The perforation holes are read-out by a laser for the most optimal flash and shoot timing, creating a digital image of every film frame. This creates a perfectly stable and balanced picture.
The laser operates with an accuracy of 0.2 mm."

Cheers!

[carllooper]

According to their web site they in fact do not utilize a sprocket based system but a combination of PTR roller tension and laser alignment via the film perfs!

film perfs = sprocket holes

0.2 mm = instability

Suppose a Standard 8 film has a 3.3 mm frame height. Divide 3.3 by 0.2 and you get a vertical accuracy of 1/16th the frame height. That's not very accurate. In fact it is so inaccurate I doubt the Muller laser has that level of inaccuracy. Perhaps it's 0.2um (micrometers) which would be accurate enough.

The real problem is introducing any dependency on sprocket holes (film perfs) in the first place.

To view the film - as a motion picture - you eventually do need to use some sort of registration method. However, it's not necessary to introduce any such method during the scan process. You select which method you want to use after the scan. For example, if you want to use sprocket registration then you can use the digital copy of those sprockets - you'll get a far more accurate digital lock on those than a laser or claw could ever get with the physical perfs.

By deferring the registration method until after the scan is done you then have the time, on the computer, to both assess the film and make decisions specific to that film, or specific to a given area of the film.

[bolextech]

Thanks Carl for your insightful comments.

In the time I was taking to research my response to Nicholas, you beat me to the punch.

Here is what I had prepared to write:

If the figure of 0.2mm accuracy is not a typo, then it certainly is not good enough for 8mm film.
Given that the height of the 8mm frame is only about 3.6mm, that's a whopping 5.5%.
Registration accuracy figures are typically 0.01 to 0.02mm for non pin-registered systems.
For example, the registration specs for the Spirit scanner are listed as ±10µm or 0.01mm.

Cheers,
Jean-Louis

[carllooper]

Registration accuracy figures are typically 0.01 to 0.02mm for non pin-registered systems.
For example, the registration specs for the Spirit scanner are listed as ±10µm or 0.01mm.

Indeed, if we take 8mm height as 3.6mm and divide by 0.01mm (pin registered accuracy) we get: 360

This means for a scan that was, say, 720 pixels high, the registration could be out by up to as much as 4 pixels (±2 pixels). If you take into account the same variation occuring in the camera that would lead to a worse case variation of up to 8 pixels (camera variation + scanner variation). And since most 8mm cameras are not pin registered you could end up with registration out by 12 pixels. And without a pin registered scanner (is there such a thing as a pin registered scanner?): the resulting variability is up to 16 pixels (±8 pixels) for a 720 pixel high image, or 1/45th the frame height, or 2.2% of the frame height.

Laser registration can improve this but it's not exact.

By using the scanned sprockets as the basis for registration you'll get an exact lock on the sprockets - meaning it will be exactly the same as the original camera was able to get (for the scanning resolution).

And using frameline recognition you'll get an even better result, because the frameline represents the relationship between camera and scene rather than between claw and scene. Frameline recognition is a little more difficult to implement of course.

With content aware registration (such as the Zapruder film) you can lock onto the scene itself - removing both camera and claw/sprocket instability. Content aware registration is actually a little easier to implement than frameline recognition because framelines can overlap causing ambiguities.