Jump to content

Not my best Super 8 footage yet.


Moises Perez

Recommended Posts

Is it possible the OP's video was re-scanned on a scanner that had a bad roller or pressure plate not right, or something the film was dragging and scrapping on as it was scanned? I would still expect the steady software to correct it, but couldn't the scanner have a mechanical flaw (bad/dry roller, dirty gate etc) along with needing better steady software?

Edited by Craig Janeway
Link to comment
Share on other sites

Is it possible the OP's video was re-scanned on a scanner that had a bad roller or pressure plate not right, or something the film was dragging and scrapping on as it was scanned? I would still expect the steady software to correct it, but couldn't the scanner have a mechanical flaw (bad/dry roller, dirty gate etc) along with needing better steady software?

 

The scanner doesn't use any mechanical registration at all. It's all software registration.

 

The scan starts off as unregistered - frames just flowing through the scanner gate however they they will. Indeed if you look at Perry's excellent download, you'll see the film is weaving around relative to the scanners gate. This is quite intentional. The film is transported in such a way as to avoid any mechanical stress on the film. Very good for transfers of old film. So the film just feeds off the rollers in any way it likes - of course it will be physically correlated to however the film was wound onto the feed reel in the first place (but we can't solve registration there!). I suspect there will be flexibility in the rollers either side of the gate to further minimise stress on the film.

 

This is the genius of software registration. It allows the physical film material to be handled very gently. Its not being forced into some mechanical straight jacket.

 

In a camera it is put through a "straight jacket". The film flows through a mechanical system designed to align the film in such a way that projectors or scanners can then decode where it was relative to the camera mask at the time of exposure. If the camera doesn't do this properly the scanner has no hope of registering the film in any generic way. It will be up to dedicated software to do tricks like matching the camera mask between frames or matching the content between the frames - neither of which works in all cases, but in the majority of cases it will. But you don't want to depend on it. For one thing it takes more computer time to do that than simply following perfs and film edges. And it would probably go bananas trying to register some of the experimental films that are made.

 

C

Edited by Carl Looper
Link to comment
Share on other sites

 

That really is brillient! Having a changing distance between edges when the software is programed to a set distance, would throw off calculations for "squareness." In fact, the slope of the change along the edge(s) may well be reflected in frame positioning via software, so we see the rotation via numbers based on a variable and not a constant, as the software is programed to expect. ... but also that the software lacks the ability to correct it self and eliminate a degrading top edge flatness. Shouldn't the software be looking at the frame and not edges of the film?

 

Dedicated software could be designed to track the camera mask rather than the edge of the film and perfs - as the camera mask is always perfect - always. BUT it depends on the camera mask actually being visible in a scan. When would it not be visible? Well, it might be totally invisible if shooting at night where only the middle of the frame had any image. Between that and a totally visible mask is everything in between, each scenario risking some possible side effect. And the computational cost is bit of a problem if your not into that sort of thing.

 

What we're trying to do here is establish a generic system, that doesn't have to use exotic algorithms to find registration. And one that can handle weaving perfs.

 

We don't expect the scanner to handle film that was incorrectly loaded. That's asking a bit much of the scanner. If the film is weaving in the camera gate that's something additional dedicated software would need to resolve. The fact that dedicated software can resolve such problems in a lot of cases is nevertheless quite cool.

 

C

Edited by Carl Looper
Link to comment
Share on other sites

  • Premium Member

So if the left edge is stationary and the right edge is weaving in this scan then it means the width of the filmstock is the variable.

I was referring to the white edges around the outside of the quicktime movie, which would be the gate in this case. It's moving all over the place and I assume that's to compensate for something.

Link to comment
Share on other sites

I was referring to the white edges around the outside of the quicktime movie, which would be the gate in this case. It's moving all over the place and I assume that's to compensate for something.

 

Ah ok. That's a really good question. I guess it depends on how close the scan was done to the film - too close and the out of bounds of the scan will make incursions during registration. The original out of bounds that the scanner would paint is probably black and we see white during neg to pos ...

 

I'm not into conspiracy theory but I'm finding I'm entertaining that side ...

 

Why is it black and white? Does the Scanstation have defaults for making positive from negative (doing orange mask removal etc)? Or is that typically done elsewhere?

 

You know there are weird details I've come across while analysing this scan that suggest this registration hasn't been done by a generic registration system. There seems to be some subtle scaling differences between some of the frames I've been looking at. Does the Scanstation do scaling as well? I can't see any reason why it would want to do that.

 

I'm starting to suspect we're not seeing what we think we're seeing. Really need to get a scan of this film where we can see the left edge and see if it really is a weaving perf or just the film weaving in the camera gate.

 

C

Edited by Carl Looper
Link to comment
Share on other sites

Why does the outside white edge of the scan move at all?

 

The large white rounded-corner edge you're referring to is the gate. When the film is passing through the scanner, by design, it's allowed to "float" inside the larger gate area. that is, the gate does not determine the edges of the scanned frame, because it's bigger than the film is wide. The reason for this is that the scanner is designed to handle film that's shrunken or damaged.

 

There is no pressure plate. There is no mechanical registration pin. It's a curved gate with a V-groove channel that keeps the film in position (plus a couple roller bearings on each side of the film at each edge of the gate, again, to keep the film in the correct position for the feed and takeup sides of the transport, but NOT to provide edge guidance within the gate itself.

 

Once the image is made, the horizontal edges (top and/or bottom) of the perforations are used to register the frame vertically. The entire frame is moved on the Y axis until the perf is lined up vertically to where it should be. The Left (perf-side) edge of the film is used as a horizontal reference point, and the film is aligned on the X axis until the left edge of the film is where it should be. Everything else falls where it may.

 

Because the overall image includes the gate, and the gate is absolutely fixed in space relative to the sensor in the scanner (that is, it doesn't float like the film does), when you fix the floating object to a point (the film on the X and Y axes), then the previously fixed objects (the gate) appear move relative to the film.

 

Of course, nobody scans with this much overscan so you would never really see that. On the OP's scan, at about 20 seconds, on the right edge of the overall frame (not the camera gate, but the whole scan), you'll briefly see the white creep into the right edge of the overall scan frame. This is the same thing we're seeing here.

 

That said, the end result is cropped to eliminate all this, so it's moot. It's just how the scanner does its registration and is this way by design.

 

-perry

Link to comment
Share on other sites

 

That really is brillient! Having a changing distance between edges when the software is programed to a set distance, would throw off calculations for "squareness."

 

...Except, that's not what it's doing. The scanner ONLY looks at the left edge (perf-side) and aligns that to a fixed point on the Y axis. The right edge falls wherever it may. The scanner does not care about that, and doesn't align anything on the right edge. Indeed, if the film is slit inconsistently, then aligning both left and right edges would warp the picture, because a right edge that's at a slight angle, that's then made vertical would stretch part of the image in the process of making it vertical.

 

but that's not what's happening here.

Link to comment
Share on other sites

Why is it black and white? Does the Scanstation have defaults for making positive from negative (doing orange mask removal etc)? Or is that typically done elsewhere?

 

Moises says "It has not been edited or processed in any form other than the color change to B&W." Presumably Moises did that on his end, but it could be done in the scanner as well. There are basic primary color grading tools in the ScanStation software that allow you to control Lift, Gamma, Gain and Saturation. When scanning color negative, the scanner is calibrated to the film base, and that process removes the orange mask. A proper flat color negative scan of correctly exposed film should require no grading in-scanner - you literally choose "no grade" from a popup menu to turn off all color correction, and just a base calibration. But in some cases, such as when dealing with a dense negative, you might have blown out highlights (base calibration sets the bottom end of the scale, putting black at a code value of 95, the standard for DPX log files. In this case, highlights might get clipped, so you'd turn grading on and pull the gain setting down from its default a bit, just to keep the whites from clipping in the highlights.

 

Of course, you could try to do a real grade in the scanner, but because it's not designed for that, it's not really advisable. You have no proper reference monitor for one, and just histograms or an RGB parade to go by, to ensure you're not crushing or clipping.

 

You know there are weird details I've come across while analysing this scan that suggest this registration hasn't been done by a generic registration system. There seems to be some subtle scaling differences between some of the frames I've been looking at. Does the Scanstation do scaling as well? I can't see any reason why it would want to do that.

The scanner is aligning the film on the left edge to a fixed point on the Y axis. If there is some slight rotation on the left edge, like if the film is going through the gate at a slight angle), it will correct for that rotation to make the left edge perfectly vertical. The entire image is rotated in that case. It does no scaling in the process of registering the frame, as far as I'm aware.
While you can set crop and scale values in the scanner, you're talking about frame by frame changes, not overall changes. The scanner only sets overall (whole scan) crop and scale values, they do not vary from frame to frame.

I'm starting to suspect we're not seeing what we think we're seeing. Really need to get a scan of this film where we can see the left edge and see if it really is a weaving perf or just the film weaving in the camera gate.

This is the point I've been trying to make. We've had our ScanStation longer than just about anyone else with one of these machines, and I can tell you that we have never seen any rocking in our scanner with any film, ever. None. never. The only time it has appeared was when the film was post-scan stabilized, as in Friedemann's footage. But the original scan had no rocking, it was an artifact of the post-scan stabilization process.
We will be scanning the film Moises posted as soon as we get it, so we'll know more then.

 

-perry

Link to comment
Share on other sites

This is the point I've been trying to make. We've had our ScanStation longer than just about anyone else with one of these machines, and I can tell you that we have never seen any rocking in our scanner with any film, ever. None. never. The only time it has appeared was when the film was post-scan stabilized, as in Friedemann's footage. But the original scan had no rocking, it was an artifact of the post-scan stabilization process.
We will be scanning the film Moises posted as soon as we get it, so we'll know more then.

 

-perry

 

I believe you.

 

The point I've been making the entire time is simply that the rocking is not in the camera.

 

So if the rocking is not in the scanner, and it's not in the camera, then it must be in some post-scan stabilisation - and if post-scan stabilisation has been done, the only reason for doing that would be if there had been sideways weaving in the camera gate - because the scanner software can't solve for sideways weave in the camera gate. Either that or it was put through a post-scan stabilisation for no reason at all (a possibility in a factory setting).

 

C

Edited by Carl Looper
Link to comment
Share on other sites

 

So if the rocking is not in the scanner, and it's not in the camera, then it must be in some post-scan stabilisation - and if post-scan stabilisation has been done, the only reason for doing that would be if there had been sideways weaving in the camera gate - because the scanner software can't solve for sideways weave in the camera gate. Either that or it was put through a post-scan stabilisation for no reason at all (a possibility in a factory setting).

 

 

There is no post-scan stabilization feature in the ScanStation. It's done in real time as the scan is being made, and once the end frame of the film is reached, the scan is 100% complete. Any post-scan stabilization is done in different software (pick your poison), and not by the scanner or its control software.

 

-perry

Link to comment
Share on other sites

  • Premium Member

Because the overall image includes the gate, and the gate is absolutely fixed in space relative to the sensor in the scanner (that is, it doesn't float like the film does), when you fix the floating object to a point (the film on the X and Y axes), then the previously fixed objects (the gate) appear move relative to the film.

 

 

That's what I thought, it's part of the "registration" function. It just amazes me how much it moves. There is A LOT of correction going on. On the machines I'm use to using, the gate never moves.

 

BTW… I was playing with the Blackmagic scanner few days ago and it's going to be out next month. It's a toy right now, but I think it has real potential in the future. It actually uses a large pixel depth imager, unlike the cameras. So right now it's setup for 4k, but they're saying in the future that may change depending on software update. They haven't quite figured out how to make it work with 16mm, but I've been told they'll have a gate sometime in the next few months. I'm trying to use that machine to transfer some stuff from a library I maintain because it would be cheap/free. So hopefully I can get some samples of 16mm eventually. The 35mm samples I scanned during the show, didn't blow my doors away in terms of registration. They can only do sprocket registration, which is weak. You can adjust where the registration is taken from, but it can only be in the area of the sprocket. So yea, for $30k it's a great solution, but I don't think it really competes against the higher end units. But it's early days and I'm certain they will do more updates over the years.

Link to comment
Share on other sites

 

There is no post-scan stabilization feature in the ScanStation. It's done in real time as the scan is being made, and once the end frame of the film is reached, the scan is 100% complete. Any post-scan stabilization is done in different software (pick your poison), and not by the scanner or its control software.

 

-perry

 

That's what I mean by "post-scan stabilisation": that which happens after (post) the scan, ie. in different software.

 

If it's not happening in the scanner (in it's software), and it's not happening in the camera (in it's hardware), then it's happening after the scan - in a post-scan step.

 

Doesn't necessarily mean it's being done by Moises. Could be additional work that the transfer shop is doing after the data comes out of the ScanStation, and before it reaches Moises.

 

 

C

Edited by Carl Looper
Link to comment
Share on other sites

The physical scan itself just captures the film where ever it happens to be relative to the scanner gate/lens.

 

The film is intentionally allowed to float in the scanner gate as part of a "be gentle to the film" philosophy.

 

What makes this "be gentle" philosophy possible is the evolution of software registration (an idea I was implementing in code in the mid 80s - to give an indication of how 'old' or 'traditional' the idea is). Instead of implementing a physical edge guide there is instead employed a software edge guide. Instead of a physical claw/pin there is instead a software claw/pin.

 

The scanner has a big eyeball (the scanning lens) attached to it's own brain (it's software) that allows it to re-imagine the floating film as if it were hardware registered. It basically does the same thing as hardware registration (that which a projector does) - but without all the mechanical vibrations, instability risks, and downright damage to the film that such might otherwise entail.

 

C

Link to comment
Share on other sites

So where are we.

 

1. There is no rocking in the camera (as very precise tests have demonstrated).

2. We don't yet know for certain if there was or was not sideways weaving in the camera (because we can't see the left edge of the film).

3. There is no sideways weaving in the scanner's software registration (there once was but not anymore).

4. There has never been any rocking in the scanner's software registration (as Perry has observed over many many scans).

5. There has been rocking created in post-scan software (as anyone who has played with After Effects and similar software can testify).

 

When one examines the magnitude of the rocking in Moises scan (think post scan here as well) and compares that to the potential sources of physical rocking (camera or scanner) the amplitude of the rocking in the scan (or post scan) of Moises film would seem to to be greater than any potential physical rocking that would occur in a camera or scanner. Looking at Perry's raw registration data one can see that any physical rocking between the scanner gate and the film would have to be very tiny. And we understand that the scanner corrects for any physical rocking anyway.

 

And insofar as the rocking in the scan (or post scan) can be cancelled out by using the edge of the film as a reference (as I've done) , and that this is also what we understand the scanner does as well, then the only place left to find the source of the rocking is in a post-scan step having been done.

 

There's no where else to go.

 

C

Edited by Carl Looper
Link to comment
Share on other sites

I've decided to make my registration test available. It starts off with two frames between which the rocking in the scan (or post scan) was found to be at a maximum (otherwise separated in the film itself by an arbitrary number of frames). So I start by cutting back and forth between the two frames, so you can see the rocking. Keep one's eye on the top right corner - that's where the rocking is most obvious to the eye.

 

I then rotated one of the scan frames until it's film edge was parallel to the film edge in the other scan frame.

 

I then used the perf to peform up/down registration.

 

I had to use the right edge of the camera mask to produce sideways registration because we can't otherwise see the left edge of the film in this scan.

 

The test ends with a long sequence of the result.

 

 

The act of aligning these frames using the same procedure that we understand the scanner otherwise uses, suggests the scanner wasn't following it's procedure, or some other software was used to do what the scanner normally does, and has inadvertently induced rocking.

 

All I've done is undo this and obtain what the scanner would normally obtain - and in the process of doing that (which would otherwise be impossible), demonstrate there is no rocking in the camera:

 

Edited by Carl Looper
Link to comment
Share on other sites

The only remaining "problem" to solve, apart from whatever mistakes were made in Moises clip, is how to properly exploit the pin-registration system of the Logmar camera.

 

As should now be obvious, a normally operating scanner obtains perfect sideways registration, and without any rocking, but there will still be a tiny amount of vertical jitter.

 

This is because the scanner uses the perf immediately adjacent to the film frame being captured, to perform vertical registration. But in a conventional Super8 camera (as much as the Logmar Super8 camera), the perf employed for vertical registration of the film, is not the perf immediately adjacent to the exposed frame.

 

Additionally, in a conventional Super8 camera the claw used to advance the film, in conjunction with a certain amount of pressure applied by the plastic pressure plate, provides for vertical registration but there's room for variation in where the perf might actually end up - as tiny as that might be. Once you factor in variations in perf pitch of Super8 film (or any other film for that matter) and that the scanner isn't using the same perf for registration as that used by the camera, there is this potential for a tiny amount of vertical jitter in a resulting scan.

 

The solution to this jitter is to use pin-registration. This simply ensures that the distance between the perf used by the camera's pin, and the exposed image, will remain a constant. A reciprocating scanner, using the same perf as the camera for vertical registration, will yield perfect vertical registration.

 

But current scanner software uses the perf immediately adjacent to the frame, producing incorrect results for both material shot on conventional Super8 cameras, and that shot on the Logmar camera. So currently there will always be certain amount of vertical jitter in a scan no matter what Super8 camera you use.

 

It is the elimination of this jitter which pin-registration resolves. The Logmar camera makes it's contribution to such a system, by employing pin-registration during exposure. And if a scanner does the same during scanning (by using the same perf that the camera uses), this will completely eliminate the tiny amount of vertical jitter that otherwise occurs.

 

One will then have a perfectly registered scan of Super8.

 

Now current scanner software doesn't do pin-registration, but there's no real reason it couldn't. But it will require some more work (ingenuity and creativity) in the scanner software, or in post-scan software to properly resolve it. The software doesn't have to do anything special - the information it needs is already there in the scans - in the position of the perf used by the pin-registration of the camera. Software just has to use that information.

 

C

Edited by Carl Looper
Link to comment
Share on other sites

 

That's what I thought, it's part of the "registration" function. It just amazes me how much it moves. There is A LOT of correction going on. On the machines I'm use to using, the gate never moves.

 

BTW… I was playing with the Blackmagic scanner few days ago and it's going to be out next month. It's a toy right now, but I think it has real potential in the future. It actually uses a large pixel depth imager, unlike the cameras. So right now it's setup for 4k, but they're saying in the future that may change depending on software update. They haven't quite figured out how to make it work with 16mm, but I've been told they'll have a gate sometime in the next few months. I'm trying to use that machine to transfer some stuff from a library I maintain because it would be cheap/free. So hopefully I can get some samples of 16mm eventually. The 35mm samples I scanned during the show, didn't blow my doors away in terms of registration. They can only do sprocket registration, which is weak. You can adjust where the registration is taken from, but it can only be in the area of the sprocket. So yea, for $30k it's a great solution, but I don't think it really competes against the higher end units. But it's early days and I'm certain they will do more updates over the years.

 

To speak of the scanner performing a "correction" is interesting.

 

Speaking in this way the film floating through the gate is a correction for the way the film might otherwise feed out of the feed reel (for example it could feed out onto the floor). And the film actually feeding off the reel is a correction for the film otherwise not doing so (sitting wound up on a reel). And the reel mounted on the scanner is a correction for the reel otherwise sitting on a bench. And the film sitting on a bench is a correction for the film otherwise in a delivery van. And so on.

 

Rather than treating all of this as a "correction" (as if there were some error somewhere) we can say it's all part of the process required to put a film onto digital screens. Rather than "correction" we say "registration" instead. We begin by positioning the film approximately - starting with delivery of the film to the transfer shop - which is an approximation of where the film needs to be, and then we fine tune this positioning: putting the film onto the bench next to the scanner, then onto the scanner, and then under the lens. Each step is reducing the "error" of where the film was previously located. Or we can say we're otherwise fine tuning the film's position. From sitting in a library somewhere on the planet to under the scanner's lens.

 

And then instead of using claw/pins and edge guides to obtain the final positioning of the film, the final positioning is done in the scanner's software.

 

C

Edited by Carl Looper
Link to comment
Share on other sites

The following is what Super8 film would look like running under the scanner lens, prior to the scanner's software registration of the film.

 

The overlaid cyan lines represent where the scanner's software registration system (using machine vision algorithms) would estimate the film should be repositioned for the final result. One can see there is some very small vertical mis-registration going on, due to the scanner's software assuming that the perf adjacent to the image is the one to use for vertical registration.

 

But in traditional Super8 cameras (such as a Leicina or a Canon) the perf used by the camera for vertical registration is a couple of frames above the gate. And in a pin registered camera(such as the Logmar) the correct perf is a couple of frames below the gate. Scanners don't use either of these perfs, so there will be some vertical mis-registration going on in a scan of Super8 film. In order for a scanner to obtain a correct vertical registration, the scanner's machine vision algorithms would need to be reprogrammed to perform some inter-frame processing of some sort (such as image stitching) in order to locate the correct perf, be it the one used by traditional cameras or by a pin-registered camera. Or the scanner would otherwise need to overscan wider, to see 3 perfs in a scan.

 

This simulation of what a scanner sees was reverse engineered in After Effects, from a scan already registered by a scanner. The original clip is that provided by Perry in a previous post.

 

Edited by Carl Looper
Link to comment
Share on other sites

Adding an eye for pin registration to the machnice vision is one option? It would only need to capture the area of perf data several perfs ahead of the the main sensor. Then, operator would assign the ratio between frame and perf, and the software would pick perf and frame relationships as it scans. A two window, or "frame in frame" monitor display can show pin registered perf and its associated frame(s) as they scan. One could mark manually with indicator mark on the first few frames to compare and see if scanner is assigning the right relationship before starting a full scan.

 

By adding a dedicated eye to the machine vision for perfs, experimentation could be done to explore if say pin registration on a 2 perf away setup, the scanner could use say the next pin registration perf one full section above what the camera use, and still get registration?

Link to comment
Share on other sites

Adding an eye for pin registration to the machnice vision is one option? It would only need to capture the area of perf data several perfs ahead of the the main sensor. Then, operator would assign the ratio between frame and perf, and the software would pick perf and frame relationships as it scans. A two window, or "frame in frame" monitor display can show pin registered perf and its associated frame(s) as they scan. One could mark manually with indicator mark on the first few frames to compare and see if scanner is assigning the right relationship before starting a full scan.

 

By adding a dedicated eye to the machine vision for perfs, experimentation could be done to explore if say pin registration on a 2 perf away setup, the scanner could use say the next pin registration perf one full section above what the camera use, and still get registration?

 

Yes, that's a good idea - would then be able to maintain a high definition capture. However such requires a hardware update to existing scanners. I guess something to do in the next generation of scanners.

 

The idea of a separate eye on the correct perf is a good idea and it's the solution I'm using in my Super8 to 16mm optical printer - machine vision monitoring of the correct perf, in conjunction with a computer driven platform that fine adjusts the camera lens in relation to the position of the correct perf.

 

In terms of scanners it's not really the scanner's "fault" as such. Many Super8 projectors won't be using the right perf either. For example, my Elmo K100 uses a perf below the gate (where a traditional camera uses one above the gate). So in terms of traditional Super8 aesthetics it's not really a big issue. It's just if we want to push Super8 in the direction the Logmar has established we'll need to push scanners that way as well.

 

In the meantime a software approach will work. I'll be writing up software to do it myself in due course.

 

C

Edited by Carl Looper
Link to comment
Share on other sites

One issue with stitching together multiple scan images (to compose a single image containing a film frame and it's registration perf) is that the scanning lens will need to be calibrated. Without such calibration there won't be a 1:1 match between across all pixels and physical space. In terms of the physical space a pixel represents, a pixel closer to the centre of the scan can differ from a pixel at the perimeter. So when stitching scans together one can get a slightly warped version of space. I found the warp to be approaching the same magnitude that an incorrect perf can otherwise provide. To counteract this one needs to apply a lens de-distorion prior to doing any stitching. So for offline processing the image stream, it would be better to get a stream from the scanner in which there was no built in registration being done, and doing the entire registration in a post scan process. A test chart in some leader can go through the scanner from which to derive tan inverse lens distortion map.

 

C

Link to comment
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
×
×
  • Create New...