Machine Vision cameras/ Mirrorless DSLR for 4K

Have to add: this is Fomapan, so good amount of grain :wink:

Hi there!

This is my first post, actually i just found this forum accidentally.
Just today i published first notes on my Telecine for 8mm based on a film projector and working with the A6300. The first examples have a lot of issues due to bathtube development, long lenses, so shaky image etc., but i really like the results, they look better then projected i would say.

Here is the post in a german forum:

And here is the first example, the quality is little shaby due to Vimeo i guess, it looks totally different from what i see on my computer.

The images produced by the Telecine are almost 6K, the 8mm fills the APS-C sensor.
I reduce it to 4K, the example you see on Vimeo is 2K.

If someone is interested in sharing ideas, i could post the description of my machine in English…

Was he using a normal projector running at 24 fps. I found this interesting video, where the poster was recording the projected screen with an iphone, and (according to him @ 30fps where the flicker disappears).

A few different thoughts to share with folks in this thread.

First, while I don’t recommend using a DSLR with a physical shutter I do want to point out that a few years back I came across a website where end-users were able to log the shutter accusations they had for their cameras. At the time, I was likely looking at models like the Canon 7D and 5D. Actuations are usually rated in the 100K to 200K range. Turns out those numbers are VERY conservative. 80% or users listed cameras as still being functional way past those metrics (500K+ if memory serves correct).

I’ll stress again that those users would not have their cameras always on or shooting the way a user here would want to.

Second, according to the famous Ken Rockwell we still have a bit of a way to go before we can really capture all the detail that’s there in 35mm film. You can read through his assessment here:

In short, 75MP-175MP might be required to truly match what our old film emulsions are capable of capturing. (So you may want to hold onto those films for future scanning).

Slightly off topic…
Last year the footage form the Apollo moon landing was digitized once again and yielded some pretty cool footage never seen before…
Historian Finds Never Before Seen Footage of Apollo 11 Moon Landing


It’s an interesting approach to analyze it, but there are some key oversimplifications he makes:

First, he states that all digital cameras use Bayer matrices and that there’s always a “lie factor”. However, if you use a monochrome camera and shoot three separate images with RGB backlights, you capture the true resolution of that sensor and thus you have no “lie factor”.

He also chooses only one film stock to calculate the overall film resolution. But the film stock he chose is on the higher end of photographic film stocks. Different film stocks are going to have more or less film resolution by a wide margin. The displacement of grains in the film itself also play a role. (If you have copies of copies of film prints, the true resolution will be significantly less.)

And there’s also dynamic range. Film cameras can have more dynamic range than some digital cameras, but by scanning the film at three different brightnesses and merging them via a tonemapping algorithm, one can get past this issue.

I recommended you take a look at this analysis; this person tests different film stocks and presents how different film stocks measure against digital cameras in both spatial resolution and dynamic range.

EDIT: I should also state that regardless, I do agree with you in that we should always keep our films for as long as possible. Mainly because they can sometimes last longer than their digital counterparts, and are useful for those continually upgrading their film scanners.

Excellent points. I’m glad someone with more knowledge than I could weigh in here.

I’m not sure how the kinograph is currently set up but I think that you’re suggesting capturing 3 monochrome images per frame which would significantly slow down the digital capture process. I’m also curious about the cameras people choose to use for this task. My guess is that we’re still well below Ken’s 75MP low watermark.

I’m in agreement about degraded film stock and poor initial quality material you mention.

I’m going to guess that within the next few years someone may well develop and OpenCV app that finds its favorite of the 24 frames and generates a composite image it likes. It may already be usable to remove spots and other artifacts during the digitization process. I’ve only fiddled with OpenCV briefly but I’ve already been impressed with what it is capable of for converting B+W images to color.

With regards to lost digital content…
I’ve lost no shortage of hard disks in my time, so I certainly feel the pain there. I’ve also had drives that somehow crush raw cr2 files. Presently I try to back everything up in at least two locations.

Yes you’d certainly be right about that, and one can compensate for this by using a higher frame rate camera, or just making the digitization process a bit longer.

I don’t think we’ll see 75MP cameras within our reach anytime soon (it’s almost twice as many pixels as widescreen 8K). Besides, the analysis that I posted estimates that the Velvia 50 stock (which Rockwell claims is 75MP) is more like 16MP in practice, which is very much within our reach. Up to a certain point, all we’d be capturing is more distorted grain and not necessarily more image detail. The only reason I’d see scanning something in 75MP is if I were scanning 70mm prints.

I highly recommend you read this thread if you haven’t yet.

Some slightly off-topic stuff here:

If you’re talking about using OpenCV to automate film dirt removal, there’s a guy who used AviSynth to automate his dirt cleaning process as well as apply a bunch of other enhancement filters.

I also have written a fully-automated experimental film dirt removal algorithm based on neural networks.

Thanks for the thorough and quality response, @ImgKD!

@WirelessGuy_NY Kinograph 2.0 (work in progress) is only capturing a single color image per frame. The camera I bought is this one:

Note that the camera can be switched out, however, and you could use whatever camera you wanted as long as it’s capable of being triggered by an external source (via GPIO pins).

After a long absence I am back with my project. I had some success with an intermittent movement based system (details later) but it is not suitable for the 35mm materials I have with torn sprockets. I now intend to experiment with a continuous film transport and would like to have your thoughts on Flea3 GigE MODEL: FL3-GE-50S5C-C: 5.0 MP, 8 FPS, SONY ICX655 as the capture camera. What are the Cons?

After a long absence I am back with my project. I had some success with an intermittent movement based system (details later) but it is not suitable for the 35mm materials I have with torn sprockets. I now intend to experiment with a continuous film transport and would like to have your thoughts on Flea3 GigE MODEL: FL3-GE-50S5C-C: 5.0 MP, 8 FPS, SONY ICX655 as the capture camera. What are the Cons?

I can’t remember off hand but isn’t GigE’s data rate slower than USB3? And do you need additional hardware to hookup a GigE camera?

With regards to the PointGrey (well technically FLIR now) cameras, the GigE cameras are at slower framerates than the USB3 cameras for the same resolution and sensor, yes.

Hi @Udayarangi - One thing I’d watch out for with that particular sensor is the relatively lower dynamic range, ~56db is perhaps a bit low when there are sensor options that approach 70+db. Although it will depend a lot on what kind of material you plan to scan. The advertised spec on the IMX265 that I’m using is 71db and I’m finding that with careful tweaking of the pipeline that it’s just enough to handle older Kodachrome.

I’m very curious and hope you share your thoughts on the continuous movement transport - it certainly puts a different set of constraints on how the film is illuminated and how the shutter is configured on the camera!

Yes, I have noticed the low dynamic range. The camera with a lens was available @ ebay for a reasonable price. (but later learned that when I have to pay for the required cables, it is not) I only wanted to test my assembly, so I needed a cheap camera with external trigger, flash connection, at least 1 fps frame rate.
I am planning to use a continues movement transport only for old 35 mm black and white materials, with damaged sprockets. I had developed a scanner using an old 35 mm projector for the prints with undamaged sprockets. (details will be published at Kinograph forum). I am going to use a Nikon J1 which I own to test, but it takes nearly 4 seconds to store the image.

Udayarangi - Nikon J1 Recording frame to a card or PC? How?

Either methods can be utilised. In camera recording is much faster but 3 seconds per frame. Externally using digicam software it takes at least more than 6 seconds to capture to the PC. What I wanted to test was the triggering.


Just a question. Can this camera be used as an “all-rounder”? For 8mm, 16mm, and 35mm? The website is in Dutch but the specs are hopefully understandable. :slight_smile:

LINK: Sharp 5M sensor CCD (2448X2048)

@Gunther_Weygers I couldn’t find the dynamic range specification for that camera. That is one important thing to look at when looking for a good camera. See johnarthurkelley’s commend above regarding good numbers for dynamic range.

Ok @matthewepler

But I’ve found something else, as you know the Point Grey camera Flea3 (FL3-U3-88S2C-C), it’s maybe an “old” camera, but for that price point, I couldn’t refuse it. It was €320 (~352US$)

And this is the dynamic range of the camera, I don’t know if this is good or not? FL3-U3-88S2C-C_Imaging_Performance.pdf (141.4 KB)

Hi @Gunther_Weygers - the Flea3 camera that you’ve got should work pretty well as an “all-around” solution. The general rule of thumb I’ve tried to follow for scanning is to get a minimum of 4K at S35mm, 2K at 16mm and ~720 lines for 8mm. This depends on how much of the film outside the actual frame area you intend on recording with your build, and as always, what you intend to do with the footage later. You may need or want more or less. The dynamic range in the high 60’s is also pretty darn good for the price.

Now that you’ve got an imager you’ve got the fun part of picking the lens!