HDR Video

gdstaples wrote on 11/21/2005, 11:17 PM
I currently specialize in HDR (high dynamic range) still photography and would like to offer these same services for my video clients.

I use a number of tools for stacking and performing image averaging for multiple images of the same scene shot at different exposures (exposed for shadows, mid-tones and highlights) and then combined in one image that often contains more than double the dynamic range of what a professional digital camera can capture. To do this I run a number of plugins and filters on the images to boost saturation, curves, levels and often have to do selective layer masking to paint in areas of blown highlights with a properly exposed image as the base image.

How can this be achieved in the video world and better yet are there any plugins that can do this within Vegas?

Thanks,
Duncan Staples

Comments

farss wrote on 11/22/2005, 3:27 AM
I've read of this technique with stills but it relies on having a camera that is absolutely locked off and it pointing at something that doesn't move, kind of contrary to the concept of 'the movies' if you get what I mean.
To come close you'd do best and cheapest shooting 35mm film and then working on a DI, expect to spend huge sums of money. As I'm certain you've come to realise one frame at that bit depth is a lot of data and you'll be storing and performing calcs on 24 of them for every second of moving image.
If you only need to make video from those HDR stills then you need to grade them into a 8 bit depth images as that's really all Vegas can handle. Even though high end systems can work at 10 bit it's pretty unlikely that the system used to view the result will do justice to it anyway. The LCD screen I'm typing this on is only good for 6 bits.
Bob.
Bob Greaves wrote on 11/22/2005, 4:11 AM
Video as a medium is not as up to that task as is film. To closely resemble the effect you would want a Hi-def camera with a large lens and an oversized 3 CCD system. The camera used by George Lucas would be good enough.

Your biggest problem is that video is displayed on a screen that lacks the potential resolution whereas film has an infinite resolution.
farss wrote on 11/22/2005, 5:18 AM
The very best digital fim cameras that money can buy don't come close to film. The problem isn't resolution or colour sampling it's latitude. Same goes for digital still camera which is the whole point of taking multiple exposures say 1 stop apart and then compositing them.
The only way to come close sort of in the digital realm is to shoot film and then scan it at 14 bits and even then the purists would argue you;re loosing a lot in the scan.
The fundamental problem is the CCDs themsleves, making them bigger helps but once you go bigger than 35mm you hit a problem with the lenses, I guess you could go to 70mm and dig out some old 70mm lenses but now we're into millions of dollars worth of custom built kit
As a side note it seem my Nikon film scanner can do a similar trick, it can run multiple scans and composite them together to bring out detail in the shadows, given how long it takes to do a basic single pass 16 bit scan I haven't been game to try this feature. A single frame at maximum res 16 bit targa file is 120MB, scanning film at that quality would yield over 1GB/sec.
Bob.
Jay Gladwell wrote on 11/22/2005, 5:53 AM

The very best digital fim cameras that money can buy don't come close to film.

That's not entirely accurate. It should read, "The very best consumer digital film cameras don't come close to film."

Actually, the large format scanning backs (used on 4x5 cameras for example) indeed rival film. These scanning backs (the top end units have over 380 megapixels in 24-bit RGB) provide eye-popping sharp focus, a wide dynamic range, total control of the image's tone range, as well as adjustment of image's density in less than 1/10th f-stop steps.

The downside is one of these units (not including the camera, lenses, etc.) is in the $20,000 ballpark! Not the kind of thing you'd buy to shoot week-end snaps.

[Edited for correction as pointed out by Kelly below.]

Too, here an interesting article that "shows" these backs can and have surpassed film.


Chienworks wrote on 11/22/2005, 6:04 AM
380,000,000 megapixels? You mean 380,000,000,000,000 pixels? That's enough for a 22,509,257 x 16,881,943 image, or bigger than 781 x 586 feet (238 x 178 metres) at 2400dpi!

I'm guessing you mean either 380,000,000 pixels or 380 megapixels. ;)

Jay Gladwell wrote on 11/22/2005, 6:13 AM

I'm guessing you mean either 380,000,000 pixels or 380 megapixels.

Yes! Thank you, Kelly, I'll correct that!!!


Coursedesign wrote on 11/22/2005, 9:25 AM
Even though high end systems can work at 10 bit it's pretty unlikely that the system used to view the result will do justice to it anyway.

This is the misunderstanding that HDRI often encounters. There is a lot of value in capturing detail in both the shadows and highlights, even as they are hopelessly far apart exposure-wise. In post processing, you then compress the image dynamic range so that you can see detail in both shadows and highlights, even though you are presenting on say an 8-bit screen.

Compared to the original HDRI image, you will now lose the ability to discriminate between nearby shadow and highlight values, but it will at least look like everything is correctly exposed.

Note that 10-bit, 12-bit and even 14-bit projectors are becoming more common, and will soon be in most movie theaters supporting the Digital Cinema Initiative in the U.S. and similar programs overseas, with 4,000 of projectors being installed imminently in the U.S.

film has an infinite resolution.

The resolution of film is limited by a) grain and b) diffusion through the four layers that make up most negative color film.

Note also that video resolution isn't limited just by pixels. Even with a $100K set of Cooke prime lenses, the resolution of a video camera is limited by the size of the CCDs, and iris diffraction on openings smaller than f/4 or so, depending on 2/3" vs. 1/3".
farss wrote on 11/22/2005, 12:48 PM
My goodness, doesn't anyone understand the difference between resolution and dynamic range?
You can add more pixels to the CCDs to get more resolution but if anything that reduces dynamic range, that's why for the same block size SD cameras in general outperform HD ones for low light.
Ever wondered why everyone makes such a big thing about lighting for video, simple answer, video cannot handle high contrast lighting. Even the very best digital film cameras don't come close to their film counterparts in this respect. Yes, resolution wise digital can give film a good run for its money and avoid a number of problems that film has but for dynamic range sorry, film wins by a very large margin, probably around 4 stops.
Bob.
Jay Gladwell wrote on 11/22/2005, 1:08 PM

... sorry, film wins by a very large margin, probably around 4 stops.

Sorry, not so far as the sanning backs are concerned. Not any more.

[EDIT]
Here's a quote regarding the latitude of the scanning backs: "High-quality digital images exhibit outstanding sharpness, clarity, and color accuracy, with a flexible tonal scale that can exceed ten f-stops when desired."


Coursedesign wrote on 11/22/2005, 2:18 PM
Sorry I don't have the URL for the following, received it in an e-mail:

Kodak image scientist Dr. Roger Morton and his team

Unfortunately there is no simple definition of "exposure latitude," but it appears that high end video is struggling to get above 10-12 stops, but this will of course get better.

One method is to have two co-located CCD pixels, one with low sensitivity and one with high ditto. This has been used in at least one still camera, from Fuji I think.
gdstaples wrote on 11/22/2005, 3:52 PM
Transparency films like Kodachrome, Velvia, Provia, Ektachrome etc., typically have around 7 stops of dynamic range.

Most lower end prosumer cameras like the 10D, D70, 20D have between 8 and 8.5 stops of dynamic range depending on how one measures DR. The Canon 1DS and 1DSMk2 have approximatly 9 to 9.5 stops of dynamic range.

An 8mp camera such as the 20D has greater resolving power than even the best fine grain 35mm films scanned at 5600ppi. One of the highest resolving consumer available films is Panatomic X at about 180/lpmm. There are very few high quality few prime lenses capable of resolving 180/lpmm and most average in the 35-50/lpmm range. If you take the best of circumstances, tripod, no shake, best quality lens etc., you can only hope to achieve something in the 140-150/lpmm range. In reality it is much less and well under 100/lpmm.

An 8mp camera and quality prime lens will out-resolve 160/lpmm. Theoretically a 35mm frame contains 16mp (mathematically) of information. In reality that number is closer to 8mp when you factor in a ton of variables. A 1DS will out-resolve any 35mm film by nearly 2:1 and the 1DSMk2 will easily outresolve the finest films by a 2:1 margin.

A 22mp Phase One P25 back will certainly outresolve a 6cmX6cm film frame and comes darn close to equaling many 4x5 films scanned at 3000ppi. The 39mp P45 back certainly equals the highest resolving 4x5 film scanned at 4000ppi and in most cases out-resolves it by at least 10%. The dynamic range of the P45 back is approximately 13.5 stops. The majority of commercially available negative films is less than 12 stops and just about every digital or photographic printing paper is less than 10 stops so the film vs. digital argument is mute here.

Back on topic:

I was just hoping that someone had come up with a way to compress a ton of DR in a scene with video much like the HDR stacking and averaging techniques.

BTW - all that HDR photography is doing is compressing a scene that exceeds that of either recording medium or output medium to fit within the color space and color depth available. Another guy did this a while back - Ansel Adams with the Zone System.

Thanks for all of the feedback.

Regards,
Duncan Staples