Home of Professional Cinematography since 1996
Published : 27th August 2003
A proper scanner does not make the compromises a Spirit/Millennium/C-Reality makes.
My own scanner, an Oxberry Cinescan 6400, has pin registration, wet gate, sequential RGB scanning to a single B&W CCD (no registration problems, no prism splitter), no compression. I can also optically zoom in or out of a frame, seamlessly and at whatever resolution I want (up to 4K).
If you do serious compositing, pin registration and perfect chroma resolution are a must. On the other hand a real scanner is less sexy than a Spirit, just like, to the untrained observer, a surgeon's scalpel is less exciting than a
Swiss army knife.
>A proper scanner does not make the compromises a >Spirit/Millennium/C-Reality makes.
Sometimes technical compromises can be very cleverly hidden.
What sort of images show up these compromises in the data telecine ?
Is it high contrast scenes or highly detailed scenes ? If I was to do a test what "type" of image on a negative would best show the advantages of a proper scanner ?
Tom Gleeson D.O.P.
Tom Gleeson wrote :
>If I was to do a test what "type" of image on a negative would best show >the advantages of a proper scanner ?
I'd be very interested in seeing "shootout" type tests between various types of continuos and intermittent scanners (like the Imagica).
>If and what do the dedicated scanners do better
It's also about what you want to do with the material once you've scannedit. Bear in mind that a 2K scan on a "dedicated" (pin registered) scanner is a genuine 2K image - a pure and simple 2048 pixel values per colour – and genuinely has whatever colour depth is quoted - 10-bit log if Cineon for example.
A datacine at "2K" actually gives you 1920 points of luminance data and only half that in chroma. In other words, only 960 photosites per colour, per line.
The compromise is justified on the grounds that that's how the eye works, with better rod resolution than cone resolution. So if you look at the untreated scans, they might look OK. If all you want is a little gentle colour correction, that's possibly OK too. (But then why go to the expense of DI?).
If you are planning any serious compositing or any significant colour correction, then you'll begin to see the limitations of the datacine files. So in other words, when you make up your comparison reel, make sure that you take both types of scan and put them both through the same processes before looking at the end results.
Tom Gleeson wrote :
>...was hoping somebody could help by pointing out the things I should >look for in a good scan ?
The two biggest differences are registration and image quality. If you need pin-registration, dedicated scanners are the way to go. For image quality, there are more trade-offs between the devices.
All telecines sacrifice some quality for speed. The greater speed allows sometimes very significant cost savings. Depending on the length of the film, a data telecine option might cost $40-60K, while the scanner option might cost $100K.
The two main difference in image quality are color resolution, and sampling structure. Most current telecines (even of the data type) were really designed for HDTV purposes. Telecine color resolution for even a 10-bit file may only have valid values between 95 and 685 [cineon] (out of 1024) code values. Sub-sampling of chroma is common. Newer systems have improved color handling and talk about 14-bit processing, but the important number is how many density values can they really read off the film. Scanners still tend to be superior in sampling structure because of CCD geometry and super-sampling. The best possible quality right now comes from a Northlight scanner scanning at 6K and down sampling to 2K.
As far as things to look for in a good scan. Look closely at high-contrast edges to see if there is any color fringing. There should be none in the middle of the frame (lens centre), but small amounts around the edges will usually be acceptable. Look for fine details, or at very tight grid patterns to see if sharpness is acceptable. Clarity and low noise in the shadow areas is also a qualitative evaluation you can make. Look at a flat field image (pixel value 100 or a 2% black card) scanned on the system and see how much noise or image structure is present. Acceptable roll off of highlights is something you always have to look for especially in telecines, but even scanners have less dynamic range that most negatives, so consider the content and how exposed it will be versus the device you are scanning on.
The difference between a scan and a telecine can be greater than the difference between release and answer print in the final picture.
so good luck...
Pacific Title & Art Studio
West Hollywood, CA
[we have Northlight and Cineon Genesis Plus scanners -- no telecines]
Jim Houston wrote :
>Telecine color resolution for even a 10-bit file may only have valid >values between 95 and 685 [cineon] (out of 1024) code values.
Although I pretty much agree with all of Jim's comments on this subject, he is incorrect in assuming that the "10-bit file may only have valid values between 95 and 685" when using a telecine as a scanner. With both the Thomson Spirit Datacine and the Cintel DSX telecine it is possible to get the dynamic range equivalent to a film scanner, i.e. black (1% reflectance) around 95-ish and white (100% reflectance) up to 1020-ish in cineon 10bit code values.
This set up of 95-1020 cineon code values are obtained from both these telecines for our DI work here at MPC. These values are obtained from the Thomson Spirit Datacine by using the CPD (Cineon Printing Density) LUT out of the Datacine and is the standard data output from the Cintel DSX.
The Moving Picture Company
London W1F 0NL
>A datacine at "2K" actually gives you 1920 points of luminance data and >only half that in chroma.
This justification of chroma subsampling is entrenched but misleading. The fovea, which receives the part of the image the eye pays most attention to, is the densest area of the retina, and is practically devoid of rods.
>This justification of chroma subsampling
Correct. Sorry - I should have put "justified" in "quotes".
My point being that even if there was better rod than cone resolution where it counts, basing the way an image is stored and processed, on the way the eye will see the final result is no "justification" anyway.