Cinematography Mailing List - CML
    advanced


HD-To-SD Better Than SD

 

Maybe some of the experts here can weigh in on this, but over at Creative Cow, we're having this discussion on whether HD downrezzed to SD is inherently better than SD origination if the end product is watched in SD.

My guess is that if both the HD and the SD cameras are 4:2:2, both have 16:9 2/3" CCD's, both have the same lens on, both shoot the same format (24P, 30P, 50i, 60i, etc.) then the HD-to-SD downrez should not be superior than the SD-originated image but look about the same. For example, if one took a 24P HDCAM recording (which isn't even 4:2:2 but let's say that it was) and downconverted it to DVCPRO50 16:9 NTSC with a 3:2 pulldown added to convert 24P to 60i, would it still be a better image than something shot in 24P using the DVCPRO50 16:9 NTSC Panasonic SDX900, which is also a 2/3" 16:9 CCD camcorder?

Someone said that the HD-to-SD downrez would still be superior because it was sampling a more detailed, higher rez image, but isn't the SD camera in this case photographing reality itself, which is even higher in detail, information, etc.?

David Mullen
Cinematographer / L.A.


David Mullen wrote:

>Maybe some of the experts here can weigh in on this, but over at >Creative Cow, we're having this discussion on whether HD downrezzed >to SD is inherently better than SD origination if the end product is >watched in SD.

David,

I am far from an expert but a good example of this was the broadcast of The Masters golf tournament today on CBS. I was watching it on a SD monitor. They were cutting promos and bumpers from the HD material and it definitely had a smoother creamier look over the pretty SD material. How much cc and manipulating they did to the HD material is anyone's guess but the 16x9 footage on a SD monitor did look different in a nice way.

Tom McDonnell
DP/Operator
New Orleans, La


>I am far from an expert but a good example of this was the broadcast of >The Masters golf tournament today on CBS. I was watching it on a SD >monitor…

But are you comparing apples to apples? Were both the HD and SD cameras 16:9? Did they both have 2/3" CCD's? Did they have comparable optics? Were they both 4:2:2? I'm sure they are any number of crappy SD cameras that of course would be inferior to an HD-to-SD downconversion.

David Mullen
Cinematographer / L.A.


David Mullen wrote :

>Maybe some of the experts here can weigh in on this…

Think of it as the equivalent of over sampling in the audio world. In a perfect camera (be it SD or HD) the MTF would be a perfect rectilinear curve from DC to the limit of the given system bandwidth. In the real world, this is, of course, not obtainable. However, since a HD camera has a higher pixel count (sampling rate, if you will, in our audio analogy) and greater bandwidth than a SD camera, it produces a flatter MTF curve in the spectrum of interest in SD and less aliasing. (There are also some colorimetry and dynamic range considerations as well.)

The net effect is that a good downconversion through something like a
Terenex can look better than an SD camera original.

Tom McDonnell


David Mullen wrote :

>Someone said that the HD-to-SD downrez would still be superior >because it was sampling a more detailed, higher rez image…

David,

By this logic film originated material on SD should look the same as SD direct originated.

I've not been to the Creative Cow website in a while, but my memory is that it is heavy on folks who come to video and HD from the computer world. This is a common attitude among computer folks working with motion images, even some of the ones making video software and hardware. For example, I often run into the idea often that if the end result is going to be a 320x240 quicktime movie, than any camera that has better specs than that will do, even an old Hi8 Handycam.

Whether you're recording sound or video, utilizing the highest quality capture device at the first stage will nearly always give better performance in the final product, even if that final product has low specs, such as streaming video.

One aspect too that they're missing in their obsession with number ratings, is dynamic range. The SD camera can only deal with so much dynamic range without losing both the top and the bottom. Both HDCam and Film are much better at compressing original scene values down into a range that can then be further "compressed" in a transfer to SD.

Now is this image difference readily apparent? It depends. Depends on how good the SD camera is. A studio Ikegami or Hitachi (Which have some pretty amazing specs themselves) is going to in a news set environment lit to the limitations of SD, give an HD camera a good run for the money. But take it outdoors, or include a window in a shot, and the difference will quickly be apparent.

Similarly, if you're shooting documentary footage, or just low budget film making with only so much time to light, the HD material should be noticeably better.

Steven Bradford
Seattle


>My guess is that if both the HD and the SD cameras are 4:2:2, both have >16:9 2/3" CCD's, both have the same lens on, both shoot the same >format (24P, 30P, 50i, 60i,)

My first digital stil camera had a couple settings. I did a test - took a picture at 640x480, then the same pic but on a higher rezzing setting which I "down-converted" in Photoshop. I just shrunk it. Comparing both pics side by side, and the one that started with more pixels was clearly better-looking.

I would think that over-sampling works like a superior compression. (I'm not sure if that's a metaphor or not). There must be a point of diminishing returns and I'm not what it is. More specifically, it would have more information of a given space to deal with and when it's rounded off, it's rounded off to a more accurate figure. Color and rez should be a bit better.

If you blew it up and looked at it, you'd see different values per pixel.

I bought a copy of Grain Surgery at NAB - saw it last year and I've been intrigued, but you take out the grain and you lose some detail and it looks pretty soft. Now, I was thinking, what if you started with HD, cranked up the gain in a low light situation (like 12db, maybe more), get your image but with noise, then take out the noise via the software, and downconvert to SD.

Would it look soft?

Dale Launer
Writer/Filmmaker
Santa Monica


>we're having this discussion on whether HD downrezzed to SD is >inherently better than SD origination if the end product is watched in >SD…

Surely down-rezzing to SD is a different process, and inherently more successful, than sampling at SD in the first place (depending on the actual process of down-rezzing of course).

Dominic Case
Atlab Australia


>HD downrezzed to SD is inherently better than SD origination

Another angle on the same thing . . .

Hasn't it been well established for a long long time that 35mm film displayed on SDTV looks better than video on SDTV? OK, some of that is the curve shape and dynamic range - but we're talking resolution here, as well.

Sampling, by whatever technology, should always be at the highest possible resolution : subsequent downconverting, reducing etc is usually a much gentler process and retains more information.

Dominic Case
Atlab Australia


>By this logic film originated material on SD should look the same as SD >direct originated.

Well, not really, because film also has other image qualities that make it superior to SD video origination than simply resolution, but it does bring up the other question, which is would film transferred to HD and downrezzed to SD look better than film transferred to SD? By what I'm hearing here, the answer would be yes.

David Mullen
Cinematographer / L.A.


> Think of it as the equivalent of over sampling in the audio world.

But bear in mind that some HD systems throw away so much original data that you may not be getting much more actual original data than you would with an SD original.

This is certainly the case in PAL…AND this reduction of original data may leave you with LESS original data in some areas, think about the colour.

Cheers

Geoff Boyle FBKS

Director of Photography
EU based
www.cinematography.net


>still more useable information under the curve in a HD downconversion >than in a conventional SD original.

Hmm,

1920 * .25 / 4.4 = 109.09

720 * .5 / 1.5 = 240

Looks like a hell of a lot more original colour data on a PAL DigiBeta than on a HDCam tape.

Actually, even the luminance is better, 480 for DB, 327.27 for HDCam.

Cheers

Geoff Boyle FBKS

Director of Photography
EU based


David Mullen wrote :

>Well, not really, because film also has other image qualities that make it >superior to SD video origination than simply resolution.

You don't believe this is also true of HD?? That the only advantage of HD is more pixels?

In the downconversion, won't some highlights that are clipped in SD, but not in HD, survive the transition to SD? They do in a film to SD video transfer.

What about the oft mentioned range in the shadows that the HDCams has? From what I've seen, that survives the downconversion also.

A good downconvertor will translate down the HD material so that it fits as well as it can within the SD space. It doesn't just "crop" out everything that falls outside the parameters of SD.

When we use film, or HD, or even just a better camera, to get to a lower level format, it is as if you've put in a compression step, not digital compression, but compression to bring the parameters of the original scene down to the capability of the SD recording format to hold it. It's like the difference between shooting a bright contrasty daylight scene with an SD Camera, and shooting it with reversal (Which will be hard to bring into SD range) or with negative, which does the job nicely.

Would the same folks who are saying that there is no difference between HD down converted to SD make the same claim that there is no difference between an Ikegami SD studio camera and PD150, both recording to the same format?

I remember playing around with this concept in the 80s, and discovering that if I was given the choice between shooting with an HL79e (The $50k) camera on 3/4" or with an ITC 730AP($15k camera) onto 1", I would choose the former. Because it would do the better job of compressing the original scene down into a good picture that would still look good on 3/4". It's much the same with HD cameras.

Geoff's point is excellent, about HDCam compression. My *suspicion* though is the HD originated material will be superior viewed on a monitor. But the DigiBeta SD original just might be better for compositing or multi-generation work? Someone else can probably speak to this better than I can, and likely has by the time I've finished this post.

Steven Bradford
Seattle


A sweet SD signal has ideally been encoded and transposed only once.

Transposition, encoding and transcoding are generally more damaging to a signal than good compression. On the flip side of the equation is the fact that the HD signal was probably acquired with a snazzy HD lens, which has a lot more resolving power than what was on the SD acquisition device.

Personal experience is that SD straight to the disk through an AJA or CineWave blows reprocessed HD/SD away (bassed on how well the skin tones hold and how well the signal pulls a matte).

Scott Billups - LA


In my experience, VariCam HD down-converted from the deck is superior to images from our Sony Digital BetaCam camera using identical lenses. We occasionally use the Sony as a "B" camera on interview style shoots so we have plenty of opportunity to see the footage side-by-side.

Granted, the VariCam footage is shot 24p and the Sony is 60i; I'm sure, as David Mullen pointed out, that accounts for a good part of the difference.

I do think that the new Panasonic 24p Pro 50 camera looks very nice, and is a bargain at under 25K, but, of course, it doesn't offer an HD field tape for future use. Since the cost difference of shooting SD or HD is but a very small part of a total production, SD acquisition is no longer a good choice for our clients.

Best regards to all,

Leo Ticheli
Director/Cinematographer
Birmingham/Atlanta


Leo Ticheli wrote :

>Since the cost difference of shooting SD or HD is but a very small part of >a total production, SD acquisition is no longer a good choice for our >clients.

Including the cost of HD decks in post?

I'm not aware of the cost of DVCProHD decks, are they now close enough to the cost of SD decks to not make much difference? I imagine you also can factor in a longer operational lifetime, and the ability to play back SD cassettes too.

On the project I shot a week ago with the little Panasonic 24frame camera we had wanted to go with an HD Varicam downconverting the HD tapes to DVCPro50, and editing that. Thus we would have been using the Varicam for it's 24 frame look, and (hopefully) closer to film dynamic range, but the Varicam camera wasn't available, and by that time we'd talked ourselves into wanting to do a 24 frame project and this was a perfect project to experiment with and take a chance on the DVX100.

(Successfully I might add…)

Steven Bradford
Seattle


The Panasonic HD decks are very inexpensive compared to Digital BetaCam; the130, for example lists for $27,000.00.

We purchased a complete system, camera, the 150 and 130 decks, and monitoring for about the price we paid for our system when we moved to Digital BetaCam.

Best regards,

Leo Ticheli
Director/Cinematographer
Birmingham/Atlanta


>whether HD downrezzed to SD is inherently better than SD origination if >the end product is watched in SD.

Back in '98 Fritz Roland showed me a demo of his side-by-sides of the then-new HDW-700 60i HDCAM downconverted to DigiBeta, and DigiBeta from whatever the top-of-the-line 16x9 DB camcorder was then. Same size 2/3" CCDs, same angles, same lighting, etc. The HDW-700 pretty much saw the tonal scale the same way the DigiBeta did (and Fritz had controlled the lighting to keep it within range for both cams), the frame rates were the same, the lenses were the same; so there really wasn't any rendering difference to look at aside from resolution.

The differences were startling. The downconverted HDCAM pix were far more detailed and realistic, making the edges in the DigiBeta original look like they had been drawn in crayon. My comparison at the time was that the DB original looked to the HDCAM downconversion about like single-chip Hi8 looked to DB.

A couple of years ago up at the Snader show I was entranced by a super detailed image playing back off a DVCAM tape. It looked WAY too good to be DVCAM original. I asked, and it was also HDCAM original, downconverted.

Supersampling really works, putting a quart of detail into a pint pot of picture. I've been steering people towards HD acquisition for an SD finish for some time now, as a way to get a noticeably better image without incurring all the costs of a full-HD postproduction process.

Adam Wilt / Camera Guy / Menlo Park CA USA


>Would the same folks who are saying that there is no difference >between HD down converted to SD make the same claim that there is >no difference between an Ikegami SD studio camera and PD150, both >recording to the same format?

Again, you aren't comparing apples to apples - those cameras don't have the same sized CCD's, don't have the same quality optics, etc. Of course a pro camera is going to deliver a better image than a consumer camera. My question was pretty specific : if you match CCD size, capture rate and mode, processing, compression, color space, lenses, will the HD-to-SD downconversion STILL look better than SD origination because it began with more pixel resolution.

Tom and Jeff's point about a reduction in aliasing artefacts makes sense to me, and the notion of oversampling and downrezing holding more information than capturing at lower rez to begin with makes sense to me too, so I guess I'm leaning towards believing that the HD downconversion will be better. On the other hand, my limited experience with shooting on the PAL Sony MSW900P didn't suggest that it had less exposure latitude than an HD camera.

Are most people in agreement that a film transfer to HD and downcoverted to SD would be better than a direct transfer to SD, as long as they were both 4:2:2 transfers, etc.?

I was just wondering if anyone's done any real-world tests. My past experience in seeing 60i HD shown on NTSC was that it looked just like Digital Betacam NTSC origination and no better, but that was not based on any scientific evidence, just a reaction.

David Mullen
Cinematographer / L.A.


>The differences were startling. The downconverted HDCAM pix were far >more detailed and realistic, making the edges in the DigiBeta original >look like they had been drawn in crayon.

I don't doubt you, but I think one trick to be solved in doing a comparison test would be to make sure that the end SD results from both cameras had an equal amount of edge enhancement -- you'd have to make sure that the HD camera and then the downconverter created a final amount of edge enhancement in SD that looked the same as the SD-originated material or else you might prefer one over the other simply because of the difference in enhancement artifacts.

(I'm reminded of this because some of my downconversions from HD have been over edge-enhanced by whoever or whatever did the downconversion.)

David Mullen
Cinematographer / L.A.


David Mullen writes :

>My guess is that if both the HD and the SD cameras are 4:2:2, both have >16:9 2/3" CCD's, both have the same lens on, both shoot the same >format (24P, 30P, 50i, 60i,)

Given the many differences between HD an SD hardware I imagine it would be difficult to do such an apples-to-apples comparison scientifically. But everyone seems to agree that downconversions, done with the best available equipment, simply do look better.

>isn't the SD camera in this case photographing reality itself, which is >even higher in detail, information, etc.?

Part of it comes down to the fact that an imaging chip is not a downconverter in the same sense. It doesn't have the "smarts" of an actual downconverter. (And of course, it's not taking advantage of the compressed dyanmic range, higher bit-depth processing, etc., of the HD signal... as has already been mentioned.)

Dale Launer writes :

>what if you started with HD, cranked up the gain in a low light situation >(like 12db, maybe more), get your image image but with noise, then >take out the noise via the software, and downconvert to SD. Would it >look soft?

The larger the image and the greater its bit depth, the less visible will be the undesirable by products of *any* type of digital processing. So in your example, if you then wanted to restore some of the lost sharpness, you'd be much better off doing it at the higher resolution before downconverting it. But to tentatively answer your question, I'd say it depends on *how* much softness the noise-removal process introduces. My guess is that even a small degree of softness (whatever that means!) will be detectable in the downconverted SD image -- in the same way that, on SD cameras, great lenses look better than good lenses, though both may have a resolving power that exceeds that of the imaging chip.

David Mullen writes:

>And I'd assume the difficiencies from a Bayer color mosaic filter design >are perhaps less when you start out with more pixels on a larger CCD?

Yes. Same idea. All else being equal it's best to start out with more pixels. One way to look at it is that with more pixels a given "deficiency" ends up being confined to a smaller area.

Four small deficiencies look better than one large deficiency that takes up the same total surface area. The simplest example of this is the aliasing (stairstepping) of diagonal lines.

Dan Drasin
Producer/DP
Marin County, CA


>…But everyone seems to agree that downconversions, done with the >best available equipment, simply do look better.

This is the kind of statement that drive me mad…"everyone knows" you mean every clown with an opinion thinks that they know.

Nobody, well very very few people, actually test for real.

I've seen tests of PAL DB to HDCam and vice versa and I could see no reason to spend the extra on the HDCam.

I have shot keying tests of DB V HDCam and the DB won! I'm not allowed to give details of these tests but you can easily do them yourselves.

I've put the numbers up, you know, FACTS! not "everyone knows"

I don't have the time to do another series of tests but I will happily do them in the future if I can get someone to donate the kit and edit facilities.

"everyone knows", not so long ago "everyone knew" the earth was flat!

Cheers

Geoff Boyle FBKS

Director of Photography
EU based


>Then, extrapolating from that, everyone should be shooting on PAL >Digibeta rather than HD and upconverting...

Agreed.

Back a few years ago, while specifying up/down converters for a fairly major installation I used a very un-scientific test to evaluate just how good various up-converters were.

I downconverted HDCAM (1080i/59.94) to 601, fed this to the upconverter and right out to a monitor. Two HD monitors, side-by-side. One looking at the "pristine" HDCAM deck's output and the other at the upconverter's output. All of the equipment was brand new, monitors included. Back then, for about$30K you could get an upconverter that would make the round trip pretty much not discernible from the original HDCAM recording. I had a half-dozen people (colorists, tape operators, engineers) try to tell me which of the two images was the original. Nobody could consistently identify the source. I thought that was remarkable. I'm not sure what it says about the original recording format though...

BTW, the source tape was a TK transfer I had done on a C-Reality.

Martin Euredjian
eCinema Systems, Inc.
www.ecinemasys.com


Sam Wells wrote :

>I doubt you could tell much difference between it and an F900 on SD >television.

This is what I'd like to know, and where the difference would be most apparent. Of course if you compare imagery using scenes lit to the limitations of NTSC, it's going to be hard to tell the difference, but what if you're shooting in doc style, or want to light more film style, that is with less fill, less gel over the windows, etc etc. Do the latest and greatest SD cameras have that same ability with Dynamic range as the HDCam?

If you're working in a multicamera switched situation, would you have to light to the limitations of the SD cameras? I know from working with the latest studio cameras in SD, that they do have amazing capability, but I don't know if they go that far.

If so, then I would have to agree, there would be little difference in the final result, to the eye. The HDCam compression might even introduce problems in the downconvert. On the other hand, if you're shooting a show that will have a lifetime beyond a couple of years, those masters will be SD masters in an HD
world.

Steven Bradford
Seattle


>greatest SD cameras have that same ability with Dynamic range as the >HDCam?

Well they've got the v3 or whatever for the F900, I don't know if they 'll do the same software trick on the MSW's - a good question. Besides, blown-out highlights on HD are worse than SD because there's more bad information.

>On the mediocre video projector at the DigiCineSummit it compared >well with the Viper -- of course one was comparing good looking >material shot with either camera, via bad projection, typical NAB kind of >event!

Sounds like the DigiCineSummit needs a Producer/Director; where's Paul Verhoven when you really need him ?

-Sam Wells


I did declare :

>downconverted HDCAM pix were far more detailed and realistic...

And David Mullen warns :

>...make sure that the end SD results from both cameras had an equal >amount of edge enhancement...

Absolutely valid point. I had no control over the cams in the tests I've seen. FWIW, the downconverted HD in Fritz's tests showed no excessive enhancement, IIRC (but it's been five years fercryin'outloud).

Almost every time I've seen "impossible" amounts of detail in an SD image, it turns out to have come from film or HD. And it correlates well with the supersampling we used to do in CGI back as Circuit Studios and how that affected the look of the final images, too, but in Geoff's favour I must add that all the comparisons I've seen have been480 (or 486) line "NTSC" vs. 1080i. Geoff's upconversion of PAL DigiBeta with its higher vertical detail may be a key element.

As to his better chroma-keying from the SD vs the downconverted HD, the raw numbers and my own experience with chroma-space interpolations leads me to believe that deficiencies in chroma interpolation and filtering are very likely the culprits here, along with a spatially periodic variation in vertical sharpness that many of the downconverters seem to suffer from. However, that's speculation, and besides, we have to work in the real world of available equipment, not the theoretical world of what *should* be. So I can't dispute Geoff's tests despite my NTSC experiences being somewhat the reverse!

Cheers,

Adam Wilt / Camera Guy / Menlo park CA USA


Talking about the IMX camera (MSW-900) - it has a 1 megapixel sensor, as opposed to the (approx.) 1/2-megapixel of other Sony SD cameras (Digibeta included). I wonder if this "oversampling" at the CCD is somewhat analogous to the original question - but instead of shooting HD tape and downconverting, it's being done at the camera head.

The MSW-900 and the HDW-730 are essentially the same camera (same basic DSP and menus) in SD and HD....and on the Panasonic side the Varicam and the new AJ-SDX900 should be quite comparable - I missed NAB this year, but I'm told that the SDX900 has many of the same menus/DSP features as the Varicam, and of course it achieves 24P the same way.

I think this Sony HD-to-Sony SD, Panasonic HD-to-Panasonic SD comparison would be enlightening.

George Hupka
Director/DP, Downstream Pictures
Saskatoon, Canada


There are way too many variables that can't be locked-down to be able to answer this question conclusively. It is daunting task to evaluate this theoretically. The empirical route is all that remains. The only thing that works is to evaluate this in the context of a specific project and configuration and pass judgement within that context. As a generalization I'm not sure that any one opinion is valid.

I have seen downconversions from telecine HD material that looks great. Then again, what was it compared to, right? An old MarkIII, Ursa, Diamond? With which upgrades? How well was everything calibrated?

Martin Euredjian
eCinema Systems, Inc.


Martin Euredjian wrote:

>I have seen downconversions from telecine HD material that looks >great. Then again, what was it compared to, right? An old MarkIII, Ursa, >Diamond? With which upgrades? How well was everything calibrated?

I suppose you could make the claim that the SD output from a Spirit or C-Reality is essentially a downconversion from HD, as you get the advantages of the higher res sensors (1920 for luminance, 1440 for chroma in the Spirit I) and their advantages in terms of less aliasing, etc. Of course, to get a fair comparison you'd have to compare the output with an FDL-60!

Jeff "used to collect telecines, but tries not to anymore" Kreines


Martin Euredjian wrote :

>The only thing that works is to evaluate this in the context of a specific >project and configuration and pass judgement within that context.

Around of the end of 2001 the production company I had been shooting running car footage for was deciding weather to shoot an upcoming project HD. They had yet to decide if they could afford HD projection for the automotive show they were producing, but wanted to keep the option open. We did a quick test, though is was not exactly apples to apples.

We had shot some HD previously at 1080i for another project. This footage was down converted at Encore in Hollywood and then edited SD on an AVID. Similar existing footage (but not shot side by side) shot SD on a Sony 600 was also edited on the AVID. The footage was output to Beta SP and was projected on a 20ft. wide screen with a Texas Instruments DLP projector (if I remember correctly.)

It was pretty clear the HD originated footage was sharper and just had a better look to it for this project. Color sampling issues were harder to judge since this wasn't a side by side comparison. I assumed that just the superior lenses on the HD camera would make for a better image, but as has been said, the variables are many.

Dan Gillman
Director of Photography
Culver City, California



Copyright © CML. All rights reserved.