RetroScan UNIVERSAL

[RCBasher]

My Cine2Digits system doesn't care what speed the film moves at, it can change as much as it likes during the scan as it is frame by frame and all I do is stamp the resulting avi with the correct (user selectable) original fps rate. Obviously there are limits, like camera frame rate, disk speed, acceptable blur if used in a CTS scanner and of course assumes no audio. If the latter, then a speed stable transport system is somewhat of a bonus! Users say they have scanned 8mm at 50fps with a 1920x1200 USB3 machine vision camera! However, once one considers the whole cycle including loading, rewinding, unloading, post and publishing, I agree with you Roger that the scanning speed is not so important as it is only part of the whole.

Frank

[carllooper]

Thanks, Carl. I will say that, while your formula is technically correct, the reality is sometimes surprisingly different. In other words, things that seem impossible on paper often work just fine in real life. When we were developing the unit, we of course captured a benchmark frame stationary and then compared all other frames to that as we played with shutter speeds and rate of travel. In theory, you would need a ridiculously high shutter speed to perfectly freeze a moving frame. But when you stop to realize there is no way to "perfectly" freeze anything that's moving, all math ultimately falls by the wayside since any formula that one might come up with still has to be imperically tested, visually, to see at what point "imperfection" degrades the image in a way that's detectable by the human eye. Without getting into specifics, let me just say that we were very surprised by just how low you could go on shutter duration and still have a viable image though we are well above that threshold.

Yes, the formulas don't take into account any perceptual factors. They describe what is otherwise happening before anyone looks at the film. If we were to otherwise take into account some viewer, the formulas would have to be elaborated. For example, if a blind person were to look at the film then we could take that situation into account by (for example) multiplying the results of the effective definition calculation by zero. The fps would then be 1 / zero (= infinity).

It's not necessary to freeze the image - even if one could. The definition of the camera makes such unnecessary. And that's what the formulas express. They allow us to calculate the permissable movement - regardless of observer. They tell us there is no gain to be had in slowing the film down below the calculated FPS, regardless of observer. They tell us it is not necessary to have the film stationary.

What an observer might tell us is if we can get away with speeding the film up. In the case of a blind observer we could get away with transferring a film at the speed of light and have the job done in a nanosecond.

But if the observer happens to be Superman (or Supergirl) the formulas don't tell us to slow the film down to a complete stop. Our anxiety might tell us that but the formulas give us a non-zero frame rate for such super observers. This is because the capture camera has a finite resolution.

C

[RCBasher]

Yes, the formulas don't take into account any perceptual factors. They describe what is otherwise happening before anyone looks at the film.

Nor taken into account the actual image data on the film. Given that the exposures will likely have been around 300 times longer than the scanning times we are talking about here, unless the camera and all the objects/subjects were bolted to granite, then again I would say all we are striving for with ultra slow scanning speeds is the film grain! Having captured the grain with a 4K camera, what are we going to do about it in order to make the digital files manageable?
Frank

[carllooper]

Nor taken into account the actual image data on the film. Given that the exposures will likely have been around 300 times longer than the scanning times we are talking about here, unless the camera and all the objects/subjects were bolted to granite, then again I would say all we are striving for with ultra slow scanning speeds is the film grain! Having captured the grain with a 4K camera, what are we going to do about it in order to make the digital files manageable?
Frank

Capturing at a resolution higher than the source image won't give you a result with any higher resolution than the source image. But what it will do, and both of these are extremely important (if not more important) and that is:

1. reduce grain aliasing
2. increase the depth of the result (more bits per resolvable area).

But by how much depends a lot on context. For my own work I've settled on a 4K workflow, but work I've done for others (and myself) have been done at 2K and were just fine.

C

[RCBasher]

Capturing at a resolution higher than the source image won't give you a result with any higher resolution than the source image. But what it will do, and both of these are extremely important (if not more important) and that is:

1. reduce grain aliasing
2. increase the depth of the result (more bits per resolvable area).

C

What's bit depth have to do with resolution? A pixel is an analogue device and bit depth is to do with how many steps we divide up the stored analogue charge.

For grain aliasing, we would be better off with a lower resolution camera and a softer lens so we do not reach the aliasing frequency. Unless you really want all the grain captured which then leads on to my point above, what are you going to do with it as 6TB for a 1hour film is not really tenable!

Frank

[carllooper]

What's bit depth have to do with resolution? A pixel is an analogue device and bit depth is to do with how many steps we divide up the stored analogue charge.

A higher resolution camera captures more pixels per area of film than a lower resolution camera. More pixels = more bits = more information per same area of film. We can describe this information in terms of both resolution AND depth. Where there is a lack of resolution in the film image, there is not a reciprical lack of depth. In terms of depth the film has a lot of it. A lot. So allocating more pixels (more bits) per area of film encodes more of the film's depth.

For grain aliasing, we would be better off with a lower resolution camera and a softer lens so we do not reach the aliasing frequency. Unless you really want all the grain captured which then leads on to my point above, what are you going to do with it as 6TB for a 1hour film is not really tenable!

This is not true. As you decrease the definition of the sensor and/or lens you increase the grain aliasing. You get a "grainier" image. As you increase the resolution of the sensor/lens you decrease this graininess.

This is because the graininess in most film scans is not the native grain of the film. It's aliased grain. Fake grain! This is a relatively recent discovery (late 90s). Back in the 60's / 70s it was completely unknown.

Aliasing is where high frequencies masquerade as low frequencies. Grain accounts for the highest frequencies in a film signal. Even if the film was shot out of focus. And grain lives above the sampling frequency. In doing so it reappears below the sampling frequency as low frequency grain. As "clumps" if you like. This clumpy grain is not in the original film. It's in the scan. The only way to alleviate this clumpy grain is to sample at a higher frequency (the higher the better). You can't otherwise filter for it because it's not a signal. Its noise. But sampling at a higher frequency will reduce the size of the aliasing. Reduce the clumpiness. It makes the graininess finer if you like.

I'm not advocating one spend money on terrbyte drives and inordinate amount of transfer times. I do it but that's just me. I'm simply talking about what I know and what I've discovered. It's not in any way changed by the size of your budget. It will still be so whether one is a prince or a pauper.

And really, if someone did a low definition scan that not only demonstrated clumpy graininess, but exploited the effect, I'd find such a work just as fascinating (if not more) as someone doing an 8K scan and exploiting the virtues of that.

C

[carllooper]

I might just add that this discussion on continuous motion capture, resolution, etc., while under the heading of the Retro, is not specifically about the Retro. It is, of course, inspired by the Retro.

Technical analysis is just a way of testing a proposed system against an ideal observer. If you can satisfy an ideal observer, you know that you can satisfy all observers (ie. including Superman etc). You then work back from that, with emperical testing, to find what would be appropriate for a keen eyed but non-super powered observer. Because going the other way (in an attempt to satisfy hyper powered observers) would be a completely redundant exercise and expense. The technical side is useful here because it provides us with a stop sign in this direction: a no-need-to-go-further sign. Because Hypergirl, as much as Superman, are fictions. Returning back from this limit and finding where to stop in relation to us mere mortals, is a lot more difficult, and it's this which requires extensive emperical testing. The technical concepts are worthless here. It is an artist's eye that becomes your fundamental tool in this regard.

The Retro itself is a perfect example of how you might balance cost and performance along the lines discussed. It is a machine of which I am quite enamoured. I've not included any notes on emperical testing of the machine (and won't be doing so) but I'm in perfect agreement with Roger's notes on this. Such testing keeps the price of the Retro down, without in anyway whatsoever compromising technical quality. And the Retro does this perfectly.

C

[RCBasher]

This is not true. As you decrease the definition of the sensor and/or lens you increase the grain aliasing. You get a "grainier" image. As you increase the resolution of the sensor/lens you decrease this graininess.

This is because the graininess in most film scans is not the native grain of the film. It's aliased grain. Fake grain! This is a relatively recent discovery (late 90s). Back in the 60's / 70s it was completely unknown.

Aliasing is where high frequencies masquerade as low frequencies. Grain accounts for the highest frequencies in a film signal. Even if the film was shot out of focus. And grain lives above the sampling frequency. In doing so it reappears below the sampling frequency as low frequency grain. As "clumps" if you like. This clumpy grain is not in the original film. It's in the scan. The only way to alleviate this clumpy grain is to sample at a higher frequency (the higher the better). You can't otherwise filter for it because it's not a signal. Its noise. But sampling at a higher frequency will reduce the size of the aliasing. Reduce the clumpiness. It makes the graininess finer if you like.

What I said/meant was true, but perhaps I did not explain myself too well. Aliasing can be thought of as a beat pattern (harmonics) which happen when any two or more frequencies come together - it is used to very good effect in up and down converting for radio transmission and reception for example. The relative spacing of the two frequencies determines the frequencies of the generated harmonics. They will always be there, it is just a case of whether these harmonics fall within our spectrum of interest or not. Your point to use a very high resolution camera (higher sampling frequency) is one method but not practical due to current performance / cost level of today's technologies. The suggestion to reduce the incoming frequencies instead by using a low pass optical filter (e.g. a "soft" lens) which will then allow the use of a lower resolution camera at the same time keeping harmonics (aliasing effects) out of our spectrum of interest. This latter approach is more deterministic because we then know what the maximum image frequency can be and limit it to match the camera, whereas with your approach we can only guess what the maximum frequency will be and take a punt with a camera of sufficient higher resolution and hope we do not get a problem down the road.

If you wish to be pedantic then regarding film speed in a CTS, Roger is correct when he says you would need a "ridiculously high shutter speed to perfectly freeze a moving frame" and your formula is incorrect because it takes no account of the vertical resolution of the incoming image. I think buckets and water make a good analogy for pixels and photons respectively: Line up a column of buckets in the yard and somewhere in the middle mount a showerhead with a fine spray above such the overall spray pattern fully covers one bucket and a small part of each bucket either side. We only have to move the showerhead the distance of a few drops of water within the spray pattern (not the diameter of the bucket) to now have a lesser fall into one of the side buckets and a greater fall into the other. So any vertical film movement relative to the image data within will be registered differentially between adjacent rows on the sensor. So the question is, what is tolerable and whilst I agree your rule of thumb formula has some merit, it does not take into account the vertical resolution of the incoming image and probably explains why the practice of higher scanning speeds relative to exposure time show no practical detriment.

Regarding your bits/pixels comments, it is generally accepted with digital cameras to talk about resolution and bit depth as meaning the pixel resolution in the X/Y sensor plane and the resolution (bit depth) in the Z plane (the ADC). They are actually the same thing, stepwise sampling of the three image dimensions - a measure of brightness at any finite point in the image and how many points in the image we measure that brightness at. I guess this is the point (no pun intended) you are trying to make. There are practical limits to how many sampling bits are required in each of these dimensions. For the ADC, we need to determine how many photons are in the pixel "bucket" (or drops of water to use my analogy). The greater the ADC resolution, the closer we get to detecting when just one extra photon falls on the sensor pixel. However, just like when emptying a bucket where there are always residual water drops left in it, pixels have a base noise level equivalent to many photons of light, so when just one more photon from our image is added it makes no practical difference to the level of photons in the pixel, so no point in having an ADC with sufficiently fine steps to detect this. Or is this true? We could take your X/Y pixel anti-aliasing solution of using a very high pixel count and apply this to the ADC to be able to detect much higher finite steps in photon count than we finally need! Guess it will depend on the number of electrons generated per photon (or photons required per electron - not looked into this!) but with sensor wells having electron counts over 32,000 then even a 16 bit ADC would only give twice the required sampling rate.

Frank

[carllooper]

This is not true. As you decrease the definition of the sensor and/or lens you increase the grain aliasing. You get a "grainier" image. As you increase the resolution of the sensor/lens you decrease this graininess.

This is because the graininess in most film scans is not the native grain of the film. It's aliased grain. Fake grain! This is a relatively recent discovery (late 90s). Back in the 60's / 70s it was completely unknown.

Aliasing is where high frequencies masquerade as low frequencies. Grain accounts for the highest frequencies in a film signal. Even if the film was shot out of focus. And grain lives above the sampling frequency. In doing so it reappears below the sampling frequency as low frequency grain. As "clumps" if you like. This clumpy grain is not in the original film. It's in the scan. The only way to alleviate this clumpy grain is to sample at a higher frequency (the higher the better). You can't otherwise filter for it because it's not a signal. Its noise. But sampling at a higher frequency will reduce the size of the aliasing. Reduce the clumpiness. It makes the graininess finer if you like.

What I said/meant was true, but perhaps I did not explain myself too well. Aliasing can be thought of as a beat pattern (harmonics) which happen when any two or more frequencies come together - it is used to very good effect in up and down converting for radio transmission and reception for example. The relative spacing of the two frequencies determines the frequencies of the generated harmonics. They will always be there, it is just a case of whether these harmonics fall within our spectrum of interest or not. Your point to use a very high resolution camera (higher sampling frequency) is one method but not practical due to current performance / cost level of today's technologies. The suggestion to reduce the incoming frequencies instead by using a low pass optical filter (e.g. a "soft" lens) which will then allow the use of a lower resolution camera at the same time keeping harmonics (aliasing effects) out of our spectrum of interest. This latter approach is more deterministic because we then know what the maximum image frequency can be and limit it to match the camera, whereas with your approach we can only guess what the maximum frequency will be and take a punt with a camera of sufficient higher resolution and hope we do not get a problem down the road.

If you wish to be pedantic then regarding film speed in a CTS, Roger is correct when he says you would need a "ridiculously high shutter speed to perfectly freeze a moving frame" and your formula is incorrect because it takes no account of the vertical resolution of the incoming image. I think buckets and water make a good analogy for pixels and photons respectively: Line up a column of buckets in the yard and somewhere in the middle mount a showerhead with a fine spray above such the overall spray pattern fully covers one bucket and a small part of each bucket either side. We only have to move the showerhead the distance of a few drops of water within the spray pattern (not the diameter of the bucket) to now have a lesser fall into one of the side buckets and a greater fall into the other. So any vertical film movement relative to the image data within will be registered differentially between adjacent rows on the sensor. So the question is, what is tolerable and whilst I agree your rule of thumb formula has some merit, it does not take into account the vertical resolution of the incoming image and probably explains why the practice of higher scanning speeds relative to exposure time show no practical detriment.

Regarding your bits/pixels comments, it is generally accepted with digital cameras to talk about resolution and bit depth as meaning the pixel resolution in the X/Y sensor plane and the resolution (bit depth) in the Z plane (the ADC). They are actually the same thing, stepwise sampling of the three image dimensions - a measure of brightness at any finite point in the image and how many points in the image we measure that brightness at. I guess this is the point (no pun intended) you are trying to make. There are practical limits to how many sampling bits are required in each of these dimensions. For the ADC, we need to determine how many photons are in the pixel "bucket" (or drops of water to use my analogy). The greater the ADC resolution, the closer we get to detecting when just one extra photon falls on the sensor pixel. However, just like when emptying a bucket where there are always residual water drops left in it, pixels have a base noise level equivalent to many photons of light, so when just one more photon from our image is added it makes no practical difference to the level of photons in the pixel, so no point in having an ADC with sufficiently fine steps to detect this. Or is this true? We could take your X/Y pixel anti-aliasing solution of using a very high pixel count and apply this to the ADC to be able to detect much higher finite steps in photon count than we finally need! Guess it will depend on the number of electrons generated per photon (or photons required per electron - not looked into this!) but with sensor wells having electron counts over 32,000 then even a 16 bit ADC would only give twice the required sampling rate.

Frank

Re. bits/pixels. Here I'm not talking about the digital side of the equation but the analog side. Prior to film exposure we can postulate a continuous signal, where the signal is not in any way quantised in terms of depth. By depth I just mean the analog "equivalent" of bits per pixel. Or the "colour" if you like. So where we might otherwise talk in the digital domain about 1 bit images, or 8 bit images, or 24 bit images (etc), the analog signal would have a quasi-infinite number of bits per pixel (so to speak). But this continuous signal is only in theory. In practice (observation) we have instead photon detections. However these detections are nevertheless statistically correlated with our otherwise postulated continuous signal. We are able to see the continuous signal despite the fact that such a signal is seemingly composed of particle detections. The pattern of particle detections induces an apparition if you like - and that apparition is of a continuous signal. Or a "ghost wave" as one notable physicist called it. And so clear are studies of this ghost wave, in controlled experiments, we are able to formulate a precise mathematical function (a continuous function) to describe it, and to which the particle detections insist on precisely conforming! This mathematical function has infinite depth. Since the observable pattern conforms precisely to the mathematical pattern, the observable pattern also has infinite depth. But the observable pattern is statistically related to the mathematical function. Or more simply, it is noisy. But as you obtain more and more particle detections the less noisy it becomes! In the limit you would make visible the mathematical function.

But the important point here is that the mathematical function describes what we see. It is not what we see that describes the mathematical function. We see a continuous signal - even in incredibly noisy signals.

Now given an area of film in which the image signal is out of focus, we will nevertheless have a signal (noisy or otherwise) which has infinite depth. So the more bits we allocate to that area the more that such will describe the signal we can otherwise see by means of looking at the film itself. One way to allocate more bits to a given out-of-focus area of film is to assign more pixels to such. The sharpness of the signal is not in any way increased (since the signal is out-of-focus) but the depth is increased. It's colour. And the noisyness of the signal actually helps to induce a more finely graded sense of the depth. Where we might have difficulty determining the colour at any single point (when masking out all other points), the moment we enlarge our mask to include the distribution of detections over an area, the more we able to see (or "hallucinate") the colour at any single point. For example we can see grey tones in an otherwise 1 bit image where each pixel is only black or white. We see the signal. We don't see the individual pixels except through an effort to do so (by blocking out all the other pixels in such an image). Film is quite nice in controlling this distribution of values, and the resulting hallucination, because it doesn't interfere with it. It has a statically neutral appreciation of the signal, which it is able to transfer to digital, ie. where an otherwise digital-only version of the same signal would produce the jaggies. In a sense film can be regarded as a very good "pre-filter" stage for a digital signal. It is able to "modulate" the original signal in such a way that helps the digital signal exhibit greater depth than it might otherwise do (had one not pre-filtered it with film).

Re. grain aliasing. This is not to be confused with signal aliasing although it shares some characteristics. Or shares some of the same theory. The important thing here is that it's a visible phenomenona, ie. not just a theoretical fiction. Indeed if the theory were incorrect (or a fiction) the phenomena would still be there. And it's cure would remain the same.

C

[carllooper]

Lets take this to another thread. I don't want anyone thinking we're talking about Roger's excellent machine. Because we're not.

C