Resolution Purists and the Real World

I love to lurk over at RedUser.net, the unofficial online forum for RED owners and enthusiasts. It’s a great place to gain insight about the technology, but it’s also just pure fun reading the various perceptions of the lesser experienced RED aficionados. The RED One camera employs a single 4520 x 2540 CMOS sensor to capture various image sizes – the most popular of which is 4096 x 2048. This is considered to be a 4K file with a 2:1 aspect ratio. Many people confuse resolution and file size, so a 4K file isn’t necessarily 4K worth of resolution. There’s also a lot of confusion between the terms resolution and sharpness. The simplest explanation is that resolution is the measurable ability to resolve fine detail, while sharpness relates to your eyes’ and brain’s perception of whether or not an image is crisp and shows a lot of detail. Both Mark Schubin (Videography magazine’s technical editor) and Adam Wilt (Pro Video Coalition) have written at length on these subjects.

 

As a poor country editor who isn’t a DP or image scientist, I defer to the authorities on these subjects, but I have spent several decades working in all sorts of image formats, resolutions and display technologies. From this experience, I can say that often the supposed resolution of the sensor, as expressed in pixels, has very little to do with how the image looks. I see a lot of folks online expressing the desire to finish in 4K, without any understanding of the real world cost or desirability of 4K post and distribution. Not to mention the fact that true 4K theatrical displays are quite a few years off, if for no other reason than the lack of financial incentive for major theater chains to convert all their 35mm film projection to something like Sony’s SRX-series digital cinema projectors. So in spite of an interest on the part of content producers to see 4K presentation venues, the reality is that high-resolution-originated product will continue to end up being viewed on various displays, from web movies to SD and HD television up to film projection and/or digital cinema projection at 2K or less.

 

 

Been There – Done That – Got the Belt Buckle

The irony of all of this is that we’ve been there before. I even have the limited edition belt buckle to prove it! In the late 70s I worked with the CEI 310 camera. This was a 2-piece electronic field camera that was definitely geared towards high-quality production and not news. The CEI 310 eventually became the basis of Panavision’s Panacam – their first foray into electronic cameras equipped with Panavision film lenses. Bear in mind that the 310 and Panacam were always SD cameras without any 24P capabilities. On the plus side, the colorimetry of the CEI camera appeared more “filmic” than its ENG counterparts, which was further enhanced by the addition of Panavision lenses and accessories.

 

At the time, I was responsible for a facility that cranked out a ton of grocery store commercials. “Painting” the camera to get the most out of tabletop shots was the job of the video engineer (often called the “video shader”). A lot of what I learned about color correction (and have since passed on to others) came from trying to get a cooked ham or roast to look appetizing using our RCA studio cameras! When Panavision set up the deal with CEI to market Panacams, they established a number of authorized rental/production facilities who would supply the camera accompanied by a trained technician. Again, this person’s job was to paint the image for the most pleasing look. Fast forward a couple of decades and you have the position of the DIT (digital imaging technician), who today fulfills the role of video shading, among other tasks, when HD cameras are used on high-budget shoots, like feature films.

 

These early attempts at electronic cinematography really didn’t go far, due to the limiting resolution of NTSC and PAL video. Sure the images looked great, but you were really only working in a medium that was acceptable for television and not the big screen. Nevertheless, companies like Panavision, CEI and other competitors (like Ikegami with the EC-35) proved that properly adjusted video cameras coupled with high-quality glass could be a good marriage, regardless of the resolution of the camera.

 

 

High Definition to Small Definition

Fortunately HD came along, reviving the ongoing interest to use electronic cameras for theatrical distribution. The company I worked for in the 90s was an early adopter of HD. We bought two of Sony’s HDW-730 cameras, which were interlaced 1080 HDCAM camcorders. Interlacing causes many of the purists to snub their noses, preferring the later 24P models as true film-style images. In spite of this, we produced quite a lot of impressive content, including a Biblical-based dramatic production for a themed attraction called “The Holyland Experience”. Our 20-minute film was shot on location in Israel and projected in a custom theater that rivaled any big screen movie theater in size and scope. The final master was edited in 1080i but encoded into 720p and projected using a Barco data-grade (not digital cinema) projector. Interlaced or not, this image was as impressive and as high-quality to the eye as if this had been a full blown 35mm film production.

 

On the other end of the scale, I’ve also posted the video portions of IllumiNations: Reflections Of Earth, Disney’s nighttime show at EPCOT – a fireworks and laser extravaganza choreographed to music. ROE’s video segments are presented on a 29’ tall rotating earth globe mounted on a barge in the middle of the EPCOT lagoon. The continental masses on that globe consist of LED displays. The final image that fills these screens is actually a 360 x 128 pixel video movie composited like a world map. The pixels for the continents are, in turn, mapped onto the matching LED coordinates of the globe. Australia only has the resolution of a typical computer desktop icon, yet it is still possible to discern imagery with a display this coarse. The trick is in the fact that viewing distances are 500’ to 700’ away and your brain fills in the gaps. This works much like the image of Lincoln’s face that’s made up of a mosaic of other images. When you get far enough back, you recognize Lincoln, instead of focusing on the individual components.

 

 

High Definition and the Silver Screen

 

Most folks now agree that the actual resolution of the RED One camera with proper lenses and accurate focus is in excess of 3K, though not quite as high as 4K. Compare this to film. 35mm negative is said to be as high as the equivalent of 8K (though 4K is generally accepted by most as “full” resolution), but typically is scanned at 4K or 2K resolution. However, the image you see in the theater from a projected release print, is generally considered to be closer to 1K. This varies with the quality of the print, projector lens and dimness of the projector lamp. Meanwhile, most of the popular HD cameras used for digital cinematography (Grass Valley Viper, Sony F900, Sony F23, etc.) capture images at 1920 x 1080, leaving you with a 16 x 9 image that’s comparable to a 2K film scan when the aspect ratio is 1.85:1. I’ve seen quite a few of the movies in theaters that were “filmed” using digital cameras (Collateral, Apocalypto, Zodiac, Star Wars, Once Upon A Time In Mexico, etc.) and I find very little to quibble about. In fact, Star Wars was shot with the wider 2.35:1 aspect, meaning that the top and bottom were cropped. So really only about 700 pixels out of the actual 1080 pixel height show up in the final prints.

 

I’ve also edited a film that was finished through a DI process using Assimilate SCRATCH. Our film was shot on 4-perf Super35mm negative and transferred to HDCAM-SR. Since we intended to end up in 1.85:1, the 4-perf Super35mm frame provided the closest fit to the 16 x 9 aspect ratio of HD, without wasting part of the top and bottom of the negative’s frame. This technique results in smaller film grain within the HD frame because more of the whole film frame is used. Internally our SCRATCH files were 2K DPX files and the output was back to an HDCAM-SR master. I’ve seen this film projected at DCI spec in the lab’s screening room, as well as HDCAM running through a projector at 1080i (interlaced with added 3:2 pulldown) and I must say that this image would not have looked any better had we worked off of a 4K film scan.

 

The reason I say this is due to the general texture of film and the creative choices made for exposure, lighting and lens/filter selection. Images that are often more pleasing to the eye are sometimes technically lower in sharpness. In other words, when you stick your nose up close to the screen, the image will tend to appear soft. Having higher resolution doesn’t matter, because there is no more real detail in the image to bring out except bigger film grain. One interesting comparison is last year’s There Will Be Blood versus No Country For Old Men. Blood went through a traditional film, rather than a digital finish, whereas No Country was completed at 2K resolution using a digital intermediate process. Both were nominated for an Oscar for Best Cinematography. By all rights, Blood should have had the higher resolution image, yet in point of fact, both looked about the same to the casual eye when seen in the theaters. The cinematography was striking enough to earn each a nomination.

 

 

It’s in the Glass

 

Going back to the Panacam example, what you start to find out is that the quality of the glass is a major factor in what ends up being recorded. I once did a film shot with a Sony F900 camera (24P). The DP/owner-operator opted to rent a “Panavised” Sony F900 (like those used on Star Wars) instead of using his own camera, so that he could take advantage of the better Panavision lenses. The result was a dramatic difference between the image quality of those lenses as compared to standard HD lenses. Likewise, some of the RED examples I’ve seen online that were shot with various non-optimized lenses, such as prime lenses designed for still photo cameras, exhibited less-than-superb quality. This is also why there have been a number of successful indie films shot with a Panasonic VariCam. Technically the VariCam, with its 1280 x 720 imager, should look significantly worse on the big screen than a Sony F900. Yet, many of these have been shot using 35mm lens adapters and high-quality film lenses. The results on screen speak for themselves. The funny thing is that there’s a lot of talk of 4K, yet when I’ve seen Sony’s 4K projector demos, the content comes from 1920 x 1080 sources – shot with various Sony or Panavision digital cameras. I can assure you that these look awesome.

 

 

You ARE Paying for Something

Aside from lenses, another thing to keep in mind is the electronics used by the camera for image enhancement and filtering. Part of the big difference you pay between a RED One and a competing Sony, Grass Valley or Panasonic camera is for the electronics used to enhance the image. The RED One generates a camera raw, Bayer-pattern image. The intent is to do all processing in post, just like sending film negative to a lab. The other cameras have a lot of circuitry designed to control the image in-camera. You may opt for a neutral, flat image, but there’s still processing applied to generate that finished RGB image from the camera, regardless of whether it’s flat or painted. This processing not only applies color matrices but also sharpens detail and reduces noise. By contrast, RED not only doesn’t apply this in-camera, but also uses OLPF (optical low pass filtering), common in digital still camera sensors. OLPF essentially filters out the highest resolution transients so that you don’t have excess aliasing in the image on things like contrasting diagonal lines, such as on a car grill. The design goal is to leave you with true and not artificial resolution. This means the image may at times appear soft, so sharpening and detail enhancement have to be added back (to taste) during the post production conversion of the camera raw files.

 

The dilemma of all of this file conversion needed in post is that you often don’t get the best results. On the plus side, you may reap the benefit of oversampling, meaning that at times an HD image downsampled to SD may look better than it if had been shot in SD to begin with. I have, however, also found the opposite to be true. HD is a very high resolution image that has more actual resolution than our monitors and projectors can truly display. An image looks more natural in HD when less detail enhancement is dialed in. If you crank up the enhancement, like you typically do in most SD cameras, then that image would look garish in HD. Unfortunately, when you downsample this very natural looking HD image into SD, the image tends to look soft, because we are used to the look of overly-enhanced SD cameras. Therefore, downsampling by a dedicated device like the Teranex Mini will give you better results than using the built-in functions of Final Cut Pro or a Kona card or an HD deck, because the Mini lets you subjectively add enhancement, color control and noise reduction as part of the HD-to-SD conversion.

 

Aliasing is another issue. A lot of HD content is captured in progressive formats (such as 24P). Progressive HD images on a native progressive display (projectors, plasmas, LCDs) look great, but when you display these same images as scaled-down NTSC or PAL on an interlaced CRT, something’s got to give. If you take a high-contrast transition, such as the light-to-dark changes between the metal bars in our car grill example, the HD image is able to retain all the anti-aliasing information for the in-between gradients in those transitions from light to dark and back. When this image is downsampled, some of this detail is lost and there’s less anti-aliasing information. The transitions becomes harsher when displayed on the interlaced SD CRT and the metal of the grill appears to scintillate with any movement. In order words, the diagonal edges of the metal grill appear more jagged and tend to “dance” between the scanlines.

 

Unfortunately this is a normal phenomenon and can exist whether you shoot digitally or on film. A few years back Cintel, an established telecine manufacturer, introduced SCAN’dAL, a feature designed specifically to deal with this issue when transferring 35mm footage to video. Although a lot of ink has been spilled about the benefits of oversampling, in some case the matching size yields the best results. I go back to SD videos I’ve cut, which were shot using a Sony Digital Betacam camcorder and am amazed at how much better these look in SD than newer versions of the same program shot on HD and downsampled for SD presentation. When downsampling is part of the workflow, then it is important to try a number of options if quality is critical. For example, sometimes hardware does a better job and at other times software is king. Some of the better HD-to-SD scaling in software is achieved in After Effects and Shake. Often just the smallest touch of Gaussian blur will help as well.

 

 

Reality Check for the Indie Filmmaker

 

One of the reasons this isn’t cut-and-dried is because camera manufacturers play so many games with the image. For example, the Panasonic HXV200 makes outstanding images and is popular with indie filmmakers. Yet it only uses a 960 x 540 pixel sensor to generate 720 or 1080 images – getting there through the magic of pixel shifting (See Adam Wilt). As good as the camera looks, when you put it side-by-side with Panasonic’s VariCam, the latter will appear noticeably sharper than the 200, because it indeed has higher resolution.

 

I’m sure you’re wondering if this is all just a can of worms. You’re right. It is. But often, the most calibrated measuring devices are simply your two eyes. Forget the specs and trust your instincts. A recent example is Shine A Light. This film was shot using a combination of 35mm film cameras and one Panavision Genesis. All footage ended up on HDCAM-SR (1920 x 1080) and the master from this not only was recorded out to 35mm film for release prints, but also IMAX. Even though HD isn’t close to the resolution of a 70mm IMAX negative, the Stones’ concert in Shine A Light looks incredible in IMAX projected onto a 5-story-tall screen!

 

In the real world, it’s amazing what you can get away with. Last year the Billy Graham Library opened with video modules that I edited and finished. The largest screen is in the Finale theater – an ultra-widescreen format that’s a horizontal composite of three 720p projections. Our sources were largely HD, but there were also a smattering of audience close-up shots from Graham’s last crusade in New York City that originated on a Panasonic DVX100A (mini-DV) camera. It was amazing how well these images held up in the finished product. Other great examples are the documentaries Murderball and The War Tapes. Each was shot with a variety of mini-DV cameras, yet in spite of the image defects, the stories and personalities are so enthralling, that image quality is the least important factor.

 

I have a lot of respect for what the team at RED has done, but I’m not yet willing to concede that shooting with the RED One is going to give you a better film than if you used other cameras, like an Arri D-21, Sony F23 or Panasonic’s new HPX3000, just because RED has a higher pixel count for its sensor. In the end, like everything else in this business, content and emotion is the most important ingredient. When it comes to capturing an image, the technical resolution of the camera is a big factor, but it doesn’t automatically guarantee the best image results from the point-of-view of your audience.

 

© 2008 Oliver Peters

Advertisements