Hello All,
This last week I've scanned my collection of 8mm film (2 7" reels, a 5" reel and about 20 3" - 50' reels). Here's an update to my previous post based on more experience and additional technical perspective. This is a long read - but it highlights important learnings when film scanning 8mm film and in particular when working with the Kodak Reels system.
With regards to optimizing capture resolution. Many who post here and elsewhere confuse actual image resolution and pixel sampling resolution - with the notion that increasing sampling resolution will increase the image resolution in the output file. I've been digitizing film (35mm slide film) for many years. To make the discussion apples-to-apples we'll move the measure of resolution to actual pixel dimensions which we can translate easily from 35mm to 8mm or Super 8. High resolution 35mm machine scans are generally 2000 dpi (25mm/2000) or about 12 micron. The reason for this stems from the actual image resolution that can be realized in color film which is about 40 line pairs / mm or 80 lines (which is easily seen on MTF charts published by Kodak and other film manufacturers). One mm divided by 80 lines is 12 micron. If you elect to increase the scanning resolution beyond this then you increase the noise in the image as you pick up more grain noise, but you pick up little or no increase in image resolution.
Turning to the Kodak Reels scanner and it's 3280x2464 sensor which is used to scan an area of about 4.4 x 5.6 mm it's easy to calculate a film pixel size of about 1.8 microns. The first thing to realize is that this already 6 times more detailed than the finest 12 micron detail in the original film! So, there is no need to pursue a higher resolution (i.e. higher megapixel) sensor. In fact, if we were to scale the sampling resolution used by 3280x2464 Kodak Reels setup on 8mm to 35mm film, we would be creating nearly 100 megapixel scans which is 2.5 times greater than even the 40 megapixel scans produced by the film industries most premiere Hasselblad Imacon 848 film scanner.
The excellent work to unlock the maximum digitization bit rate is actually a direct consequence resulting from the high oversampling of the 3280x2464 sensor. As mentioned, sampling at this higher resolution, at pixel dimensions smaller than the actual detail found in the film, results in an ever increasing amount of image noise. This increase of noise creates the need for an increase in bit rate to digitally record the noise. When the bit rate is insufficient, then compression artifacts are introduced with increasing frequency and magnitude. Unlocking the set bit rate factor in the FW - from 6400 to 15420 or so has increased this headroom by a factor of 2.5. I've taken this one step further but REDUCING the output frame size to 1440x1080 and set the frame to capture just the actual image frame. This maximizes the effective bit rate (i.e. bits per megapixes in the frame) and also creates a standard image size (i.e. 1080p) which gives me access to tools which don't incur come at a higher price tag moving above 1080 towards 4K. The 1440x1080 size translates to a 3 micron pixel size in the original film with is still 4x more detailed that the smallest detail that the film was able to record during the original filming.
With this as background, here is how I now optimize my scanning. First, I wanted to revisit the Sharpening setting, to ensure that I'm capturing as much detail as possible. As has been previously posted, reducing the Sharpening setting reduces "artifacts" with the suggestion to work at a setting of -1.5 or -2.0. I too found this to be true, but closer investigation revealed as one would suspect the Sharpening setting is a slider which sharpens on one end and blurs (denoises) on the other end. I found that ultimate sharpness decreased at a dropped down to a setting of -2.0 and possible also at -1.5. I also found that there maybe some attributing of image noise to compression artifacts. It is quite clear that a reducing the Sharpening setting to the lowest values is denoising the image.
Digitizing film has always been an exercise in maximizing image detail while removing image noise. Of course, the amount of noise one might like or wish to remove is a matter of individual preference and there is usually some loss in detail. Traditional drum scanners used in the film industry would open the actual sampling aperture typically to 12 microns so as to average out variations in noise below the actual image resolution. When film is a scanned at higher resolution (i.e. at pixels sizes less than 12 microns) then digital methods apply. For still images techniques are limited to spatial noise filter. For movie film temporal filtering is include and essential. There is a huge variation in capabilities and performance in the various software tools available and the marketing information available is almost useless in pointing one towards a best solution. Getting a top performance tool directly affects the quality of the final video as well as how the settings (in particular the Sharpening setting) is set in the Kodak Reels scanner.
After quite a bit of searching and experimenting using both freeware, online paid services and also purchased noise reduction software tools and visiting with a leading professional film digitizing service, I found a solution that achieves the top image quality I could find, together with affordability ($85 one-time perpetual license), processing speed (about 12 fps) and my existing hardware/software compatibility needs (Win 10). Specifically, I've moved to Neat Video - Home version ( see
https://www.neatvideo.com/ ). The product they supply comes as a plug-in with flavors for virtually any of the leading video editing software platforms. In my case, I use Davinci Resolve - the free version. The software also makes use of my computer's graphics card as hardware acceleration.
I captured film clips from Kodak Reels using Sharpening Settings from 0.0 down to 2.0 and after creating a representative noise profile and filter processed and assessed them for maximum detail and overall image and playback quality. The results clearly showed increased image definition with Sharpening set to -1.0 or -1.5. With a setting of -2.0 image sharpness decreased. At values of 0.0 and -0.5 there was less overall image quality due to increased noise to be treated. In the end I have opted to use a Sharpening setting of -1.0 for all scans. The reality is that I want to apply the minimum amount of image processing possible in the Kodak Reels scanner where processing power and algorithmic performance are both limited - leading the artifacts we're trying to avoid, including minimizing any image detail.
Of course, sharpness is only one factor towards getting an optimized output file. In Davinci Resolve I am also able to solve the film speed issue (16 vs 20 fps - set Speed % to 80 with Ripple Timeline enabled) and I applied image stabilization (Similarity mode, Zoom enabled and Cropping Ratio set to 0.90) - both fully supported in the free version of Davinci Resolve. In addition to denoising, the Neat video plugin also does and excellent job of natively removing Flicker (which is part and parcel to film digitization), as well as Jitter, and Dust and Scratches removal. Putting all this together creates highly viewable video files. After now watching hours of raw and processed files, I find that I am seeing new tidbits - details in the movies that tell more of the stories this movies captured years ago. My explanation for this is simple. Instead of my eyes, neural pathways and having to sort through shaky noisy, flickering, dusty and scratchy videos, it sees the nuances of the past and puts a big smile on my face.
Finally here are two important hardware tips and my positive observations.
- I too have chipped off the glue and manually focused the lens. I've also beveled the tabs on the camera cover so that I can pull it on and off without having to open the unit if I want to recheck / readjust the focus.
- Also before each new recording, with the display on showing real time the recording frame, but before I load the film in place, I check that there is now spotchiness. If there is then I take a cotton-tipped swap and clean the backlit white illumination screen. Sometime I find it helpful to dampen the swap with isopropyl alcohol.
- I've had excellent reliability. I've yet to have a splice or any other point in the scanning process get jammed in the scanner. I have had one splice open after it passed through the scanner head on the way to the take-up reel, but I can't attribute this to a design issue in the product.
Below is a raw capture (of me and my Grandpa in 1958) and then the fully processed file for comparison. Do make sure that you select 1080 HD playback when you view.
Raw
Processed
