This article will appear in the August 2001 issue of DV MagazineCOVER PHOTO
This article will appear in the August 2001 issue of DV Magazine
COVER PHOTO
Factory presets represent the "safe zone." Art isn't safe.
What we've got here is a new art form, not just a new brush. The thing to keep in mind is that the people who made the brush, whether Sony, Panasonic, Phillips, Ikegami or JVC, made a very good, middle of the road brush, with the finest components available.
Let's face it, you use a Sony DigiBeta or a Panasonic DVCPRO and you can rest assured that basically anything you shoot is going to look really good when broadcast. The cameras were made for broadcast, the factory presets are all dedicated to jumping through SMPTE hoops and I've yet to hear a dissenting voice with regard to the signal integrity of any of them in conventional video production.
HD on the other hand, especially 1080/24P is an image spec that is by its very nature, destined for a myriad of distribution modalities. Are you shooting your HD for Network broadcast, or satellite distribution in MPEG, or DVD-ROM, or perhaps a theatrical release is in your plans? If so, are you planning on printing your HD to film using an intermediate stock on a Laser printer or camera stock using a Celco? Maybe you're planning on a digital projection. Will that be using a DLP, LCD or light valve technology? So many choices and every single one of them requires a different kind of signal. A different pallet, if you will.
The HD equipment manufactures are all faced with the same problem; with HD being distributed in more than a dozen different ways, what constitutes an appropriate factory preset?
Once on tape, I imported the elements into a Mac 500x2 using the Pinnacle CineWaveHD system, and created my effects in Adobe AfterEffects5.0.
Obviously, the first step in evaluating anything is establishing a quantifiable point of reference.
Now there's a lot of charts out there to be sure, but the ones that seem to be the almost unanimous favorites are the DSC - CamAlign series and the ever popular McBeth. I've yet to see an HD shot that started with them go to far wrong. The nice thing about the DSC charts is that they transpose well on both the waveform and vectorscope, provide great density calibration and also serve as a 'min/max' for effects and printing reference.
The first pass was using 'raw' image with no Matrix preferences. What they found was that they could only over expose 1/2 stop and under expose 1 1/2 stop before the signal started getting noisy. The SMPTE-240M setting expanded that to 3/4 stop over and 2 stop under while the ITU 709 setting allowed them to over expose a full stop and under expose 3 stops before noise was introduced.
Wexler has their own preset which starts with ITU 709 but desaturates color and stretches the blacks. Sean was quick to add, "The 'Wexler Look' gave us the same room to move the color around but it didn't need as much timing on the diVinci because the look was already there. Auto Knee or DCC Dynamic Contrast Control really helps pull back over exposure in all cases."
Anxious to "complete the loop" in his understanding of HD for film, Sean was now setting the lights for yet another test.
The image that was coming directly out of the F900 via the HD-SDI connector looked crisp and full on both the scopes and NTSC monitors, but seemed a bit flat when we brought it up on the computer monitor as the shot to the right illustrates. Bright on the Mac, dark on the NT and neutral on the SGI. Keep in mind that this image essentially represents only the factory preset.
While Sean checked the connections and squirted a signal into E-Film's data stream, Jeff plopped back on a couch and twisted open a Pellegrino. "We're trying to map the base line response of the camera through the E-Film system so that they can calibrate their system for a one-to-one transfer" he offered.
We talked a bit about the potential for increasing the F900's latitude when, in a hushed conspiratorial tone, Jeff leaned forward a bit, "In conjunction with Disney Secret Labs, we've stretched the gamma and shut down the knees to the point where we had to turn the matrix off because the colors were becoming oversaturated. Since the camera is inherently so quiet you can stretch the gammas enough to put them into their linear response zones. The sensitivity that you get is just phenomenal, the camera was rating around a thousand ASA, and that's with a 180 degree shutter."
I had heard the rumors about how Sony now has the means to run Cineon colorspace on the F-900 camera using custom gamma curve software. Both the camera and a laptop need to have new software which is in limited distribution with Lucus Film and Manix Entertainment. I ask for a copy of the software and was politely turned down, but the essence is that you derive the gamma curve on your laptop and then burn it into the camera via a Sony memory stick.
Bill takes the obligatory look through the camera's viewfinder and then adds, "If you want to map the colors that the camera's sees over the full color gamut of film, you can maintain the gammas so that the gray scales match and then highly saturated colors should map out to match the color profiles on the film stream." In this case, the methodology of shooting WISIWIG (what you see is what you get) for the monitor would need to be amended because you'll actually get more color back from the film than you see on the monitor.
Sean adds, "Factory preset on the F-900 generally sets the knee point between 95 to 98. Because of the greater latitude of film we've moved the clip point to 106." Jeff Cree is quick to point out, "We've changed the slope angle slightly so that we can see three shades of gradation and a stop and a half over above 100%. That's the advantage of the 900 where you've got 23 samples between 100 and 105 compared to about five samples with the 700A. Of course people will still need to experiment and test before using any type of expanded setup in production."
My brain was just about full so I excused myself and headed back to the I/O to pick up my drive. Before leaving I transferred both the raw and the timed image sequences onto my FireWire drive as uncompressed RGB files. Once home I loaded a few frames as data files into one of my Mac Workstations and took a look at the gamma and histogram settings.
To the left are the histograms from four different software applications all displaying the levels of the painted chart that Jeff and Sean shot.
At the top is the histogram from Discreet's Combustion and then immediately below on the right is Adobe AfterEffects5. The chart on the far left center is Adobe PhotoShop and then below that the Pinnacle Commotion chart.
I'll use Adobe PhotoShop's Histogram for this demonstration, but all professional graphics applications that I am aware of use this metaphor.
If you download the actual frame, import it into any professional graphics program on any computer platform, the histogram will be identical.
I am now assuming that everyone is on the same page, more or less, so in the interest of space, I'm cropping the extra buttons and windows out of the PhotoShop Histogram, but the data profile remains intact.
With an HD signal that is destined for NTSC this is not such a big deal because parameters for broadcast greatly truncate the edges of the 'voltage potential.'
I don't really know if anyone's ever rated the histogram this way but I've inserted (in red) where I think the F900's center knee articulation is (K) as well as where I consider Ped black (B). Even though the F900 is essentially a computer peripheral, It does not record computer black.
Since it is fair to assume that they brought the knee up near 90 and ped down to 0 the resultant histogram could be considered an optimum production setting for a wide variety of uses. Maybe a little punchy for NTSC, but definitely a better signal for printing to film.
I urge you to try this for yourself. Simply download the raw HD frame, (if you have small pipes drop me a note and I'll send you a smaller version) and import it into PhotoShop. In the menu go to 'Image' and then scroll down to 'Adjust.' Select 'Levels' from the pop-up menu and then slide the black and white triangles in to the foot of the image data. Save the image as another name and then open up the histogram again. What you'll see is a signal that encompasses a wider swath of the histogramic range, but is noticeably missing entire frequencies of spatial data.
Keep in mind that this comparison is solely designed to show the potential resolution differences between timing in post vs. timing in the field and there are numerous considerations that must be taken into account before embarking on any radically new approach. Another Santa Fe Workshop regular is Charles Caillouet (C.R.) who points out that "There are pads at the top and the bottom of the HD dynamic range, one put there originally in the CCIR 601 spec in the 80s to allow for proper filter operation. There is also some pad in the camera design at nominal settings. You can safely recover some of this range in the field by careful setup, ala Jeff Cree. In post, you can regain the rest BUT you may introduce edge artifacts down the line if you don't leave the room at the top and bottom of the range for proper filter operation."
"Digital filters, like analog ones, need a little space for overshoot to retain enough information to make the signal fully recoverable. Otherwise, if you just clip the signal and throw away the overshoot, you may generate a nasty "video-looking" edge down the line when a reconstruction filter tries to use the data that you have now synthesized."
C.R. points out " One of our nasty video artifact problems is caused by saturated colors hitting the limits in red or blue channels and not being very obvious in the luma display. A gamut limit warning would clue us into that condition."
It seems as though a simple addition to the user presets that allows us to switch the camera's response profile in the field to more closely match the distribution mechanism for the project would give us as much as 20% more usable resolution.
If by some quirk of fate you're still hanging in there, I've got just one more thing to add. The graphic on the right represents the first DSC chart in this article as acquired using an Ikegami HDL-37. The signal was timed and acquired by my CineWaveHD system. It never touched tape and was recorded at the fullest bandpass I could subjectivly adjust it to.
When printed to film and projected, it qualitatively blew the other charts away. Not just to my eye, but to the eyes of directors, producers and people who print film every day of the week.
Yes the CineWave system is 8bit, and yes the HDL-37 is an interlace device, although many visual effects people (myself included) shoot composite elements at 60i simply because there is more data to work with. There are a hundred little variations on any approach we take, but at some point the discussion ends and a new one starts. It starts with quantifiable data, experimentation and testing.
Scott Billups is an LA based filmmaker. His latest book, Digital Moviemaking is available everywhere.
Just about anyone who's actively involved in HD will agree that the Advanced Digital Cinematography Course at the Santa Fe Workshops (www.snatafeworkshops.com) is ground zero for HD training. I recently met up with two of it's leading instructors, Jeff Cree and Sean Fairburn at E-Film where I was printing some of the effects I'd done for David Lynch's latest motion picture MULHOLLAND DRIVE. The movie was shot conventionally on 35mm film but I had telicinied the 22 effects shots to 1080/24P-D5 at Encore. 
While I was in I/O getting my data files downloaded, Jeff and Sean were in one of the adjacent HD bays setting up a Sony F900 which they were patching directly into the E-Film print stream. The first thing they did is calibrate the camera to factory presets while focused on the DSL chart. 
A few weeks earlier, Sean and Steve Lucas from Wexler had taken an F-900 into the telecine bay at WestWind Media, framed up a DSC chart and ran several of the cameras preset looks through the daVinci. Sean notes, "We wanted to see if a neutral camera setting or paint job limits the options in post."
I snagged a few seconds of image off of their data stream and stashed it away in my client folder. I continued to watch their progress from the I/O bay occasionally popping in to chat and watch the two HD bulls rummage around in the tiny cubicle. 
Then the shot was 'painted' by Sony's Jeff Cree (indisputably one of the most technically savvy HD engineers in the industry) and lit and operated by Sean Fairburn (one of our top HD/engineer/operators) and there it was, the theoretically perfect, HD-for-film signal.
Bill Fightner, the CTO of E-Film stuck his head in the door so I immediately ask him about printing the expanded range HD images to film. "Both Camera stock and intermediate stock have the potential to make your HD look stunningly cinematic" he offers. "While printing HD to a camera stock guarantees you the addition of grain and the cinematic nuance of commonly used film types, the intermediate stocks from both Kodak and Fuji offer a slightly wider range of colors. In some instances, the colors in your HD data stream might simply not be available in a camera stock."
Think of the histogram's box as representing the potential bandpass of your signal. Far right representing white, far left representing black and the vertical dimension representing the number of pixels of that density and color.
(keep in mind that there will be the tiniest variation due to JPEG compression)
To the right is the histogram of the raw image that Jeff and Sean shot using the F900's factory presets. Notice how nice and 'middle-of-the-road' the image is. Not just in the visual spectrum, but all the way down to the binary essence of the signal. The space to the left and right of the data group represents wasted bandpass, in this case more than 20%. 
To the immediate right is the histogram after Jeff Cree painted the F900's signal using the WFM. Notice how much wider the signal is. It has significantly more quantifiable dynamic information in it than the factory preset signal. While you might not see the difference in NTSC, I've printed these to film and there is a notable variation.
This is the histogram of the factory preset signal after it was 'timed' in post. Notice that the envelope is almost exactly the same as the painted histogram but the white vertical bands represent missing data. The Histogram is not an approximation, it is an exact reproduction of the data that it is fed. By timing in post you get a signal that looks better to the eye but is technically inferior.
The question now rises, why are we so predisposed to use the analog world standard waveform and vectorscope when we should be using a more digital friendly system such as the histogram and correction curve? Truth is that they both give us a visual representation of the line voltage potential of our signal components. A tiny AJA or Miranda converter can quite easily get a down converted signal into most laptops where you can run it through a variety of image analysis tools.
(A sample chapter of his new book can be downloaded from his website at www.PixelMonger.com)
