4k vs more DR/color resolution w/ 1080p....moot point?

[hojomo]

There's been quite a bit of back and forth on this subject in various threads. I don't have the greatest understanding of color resolution in video/codecs, the surrounding variables and technology that delivers it etc...I have recently jumped into the video world as a longtime photographer...

Anyways, what occured to me is to ask this question: does any given -- say APS-C sensor for sake of argument -- camera system that can deliver 4K 24p 8bit 4.2.2 (at a decent bitrate/codec to allow for grading), not also be capable of higher color resolution/bitrate 1080p 24p? Isn't it ultimately just a matter of having the overall processing capability and software needed to produce any given quality of video (framerate, pixel resolution, bitrate, color resolution) from the raw sensor data?

It seems like 4K vs better 1080p is cast as being mutually exclusive in a lot of the comments. I can imagine a myriad of reasons any given camera manufacturer might or might not allow for both, but would appreciate a clarification if my assumptions here are correct or not.

I've also read a few times where someone mentioned downscaling 4K video to 2K increases the color resolution...is this correct? It seems to make sense to me that sampling would increase but not bit depth....so what does that really give you? I'd have to admit that I still don't quite understand how color bit depth relates to color sampling (http://www.dvxuser.com/articles/colorspace/) in video. Are they related or entirely separate?

Bit depth makes sense to me coming from the photography world, color sampling still is a murky one

 

[combatentropy]

does any given -- say APS-C sensor for sake of argument -- camera system that can deliver 4K 24p 8bit 4.2.2 (at a decent bitrate/codec to allow for grading), not also be capable of higher color resolution/bitrate 1080p 24p? Isn't it ultimately just a matter of having the overall processing capability and software needed to produce any given quality of video (framerate, pixel resolution, bitrate, color resolution) from the raw sensor data?

Yes.

It seems like 4K vs better 1080p is cast as being mutually exclusive in a lot of the comments. I can imagine a myriad of reasons any given camera manufacturer might or might not allow for both, but would appreciate a clarification if my assumptions here are correct or not.

Pixel count and pixel size are at odds with each other, as you would guess. More pixels bring more detail. But bigger pixels bring (a) more dynamic range and (b) better light sensitivity. It's a balance. I've posted enough on DVXUser, and questions return so often, that I can now just start posting links to my answers in other threads. Here ya go! http://www.dvxuser.com/V6/showthrea...raw-the-line&p=2460586&viewfull=1#post2460586

I've also read a few times where someone mentioned downscaling 4K video to 2K increases the color resolution...is this correct?

Yes.

It seems to make sense to me that sampling would increase but not bit depth....so what does that really give you?

More accurate color.

I still don't quite understand how color bit depth relates to color sampling

Depth is how many colors you have to paint with for the whole picture. 8 bit means you have 16 million colors. 10 bit means about a billion. 12 bit means about 69 billion.

Sampling is whether each pixel can have its own color or if pixels are grouped into blocks of 2 or 4 that have to all be shades of the same color. http://en.wikipedia.org/wiki/Chroma_subsampling

 

[Samuel H]

A 4K image with 8-bit 420 color has enough information to give you a 2K image with 10-bit luminance and 8-bit 444 color.

 

[stokstad]

I've been much more interested in an 2K 10bit image with a decent but size-efficient codec, nice motion and clean blacks than 4K. But if 4K would give you all this color-information it could be a alternative to downscale i suppose. I also fear that early 4K cameras with compression codec will deliver very smeared motion and that all this information will not really be consistently captured.

Also, the look of 4K seems to be an aesthetic choice even if you downscale. I wish we would get something like the D5300 image in a FF-sensor with a slightly better codec before everyone goes 4K, it would be really great for my type of work.

 

[joe 1008]

I'm somewhat intrigued with the possibility to get kind of a 10bit HD image out of a 4K 8bit image. It might be as simple as converting H264 HD files into PRORES with the extra benefit of using more of the codec's possible colour space.

 

[Razz16mm]

I'm somewhat intrigued with the possibility to get kind of a 10bit HD image out of a 4K 8bit image. It might be as simple as converting H264 HD files into PRORES with the extra benefit of using more of the codec's possible colour space.

You have a little more room to manipulate color converting to Prores, but highly compressed codecs like 4:2:0 H.264 are very good at throwing out every bit of data not needed to represent the image as it was encoded. There is no extra data there to work with. You start pushing color gamut and you quickly get ugly artifacts like color bleed and macro blocking.
Oversampling 4k for HD can give you a richer gamut within the limits of REC709, but you have to shoot to at least a 10 bit 4:2:2 codec at 50mbps or greater to gain any real advantage in post.
Nothing beats uncompressed raw for gradable color gamut.

 

[Thomas Smet]

The 10bit may be a bit of a stretch but you really do get much better quality 1080p from a 4k source.

1. 4k 4:2:0 scaled down to HD gives you a close to 4:4:4 1080p. It isn't a perfect true 4:4:4 due to how pixels use surrounding pixels to calculate a value but it is pretty darn close. Especially if you smooth the chroma first like After Effects or 5DtoRGB does.
2. Compression artifacts tend to disappear when you scale down an image to 1/4 resolution. So for example a highly compressed 4k video scaled down to 1080p would eliminate most of the artifacts and blocking. 2x2 macro blocks get canceled out, 4x4 macro blocks become 2x2 blocks and so forth. You really do end up with a 1080p that looks like it was recorded at a very high bitrate.
3. Alias and moire is for the most part gone when shooting 4k. This then allows the users to down scale to 1080p which will typically give a much cleaner down conversion than the binning process in most DSLRs. Even if DSLRs added a higher quality 1080p recording mode they would still be plagued with aliasing and moire.

 

[joe1946]

Color: Ultra-Color Spectrum and True 10-Bit Color

VIZIO’s Ultra-Color Spectrum fully supports true uncompressed 10-bit color for the most life-like colors ever seen on a TV. Unlike most TVs, which can only reproduce a conventional color gamut of Rec 709, Ultra-Color Spectrum in the Reference Series widens the color gamut, rendering colors closer to a level the human eye can discern. Full 10-bit color enables over one billion color shades for the most life-like imagery. Collectively, these technologies push the boundaries of color and contrast to create an entirely new level of realism.

Contrast: Active LED Zones x 384 and High Dynamic Range

Along with the ultra-bright, 800 nits backlight, the Reference Series features Full-Array LED backlighting with 384 individual local dimming zones known as Active LED Zones, which dynamically adjust to match on-screen content for the deepest, purest black levels and highest contrast. VIZIO was one of the industry’s first to implement Full-Array LED backlighting in its TV collections, and over the years has continued to invest in the technology to advance local dimming capabilities to new levels of performance and definition. For 2014, all VIZIO TV product lines will support Active LED Zones, with the Reference Series serving as the pinnacle of local dimming performance with 384 zones. In addition, Active Pixel Tuning delivers intelligent brightness adjustments at the individual pixel level for increased picture accuracy and contrast.

Analyzing the lightest and darkest areas on the screen and extending the dynamic range without loss of detail in bright or dark areas, High Dynamic Range creates a contrast range with true-to-life intensity, more accurately reproducing the nuances of the picture and revealing fine details found in real scenes. Working together with an Ultra-Bright LED Backlight of 800 nits, the Reference Series delivers an unprecedented range of contrast with nearly twice the luminance of standard HDTVs.

http://store.vizio.com/news/vizio-u...performing-tv-collection-the-reference-series

 

[Razz16mm]

The 10bit may be a bit of a stretch but you really do get much better quality 1080p from a 4k source.

1. 4k 4:2:0 scaled down to HD gives you a close to 4:4:4 1080p. It isn't a perfect true 4:4:4 due to how pixels use surrounding pixels to calculate a value but it is pretty darn close. Especially if you smooth the chroma first like After Effects or 5DtoRGB does.
2. Compression artifacts tend to disappear when you scale down an image to 1/4 resolution. So for example a highly compressed 4k video scaled down to 1080p would eliminate most of the artifacts and blocking. 2x2 macro blocks get canceled out, 4x4 macro blocks become 2x2 blocks and so forth. You really do end up with a 1080p that looks like it was recorded at a very high bitrate.
3. Alias and moire is for the most part gone when shooting 4k. This then allows the users to down scale to 1080p which will typically give a much cleaner down conversion than the binning process in most DSLRs. Even if DSLRs added a higher quality 1080p recording mode they would still be plagued with aliasing and moire.

The problem is not the 4k downscale, the problem is 4:2:0 AVC encoding throws away 90% + of your data. You don't get the benefit of oversampling in the color data. You do get better image resolution and cleaner images with fewer artifacts, but it won't help the color gamut and it doesn't give you more grading room.

As far any form of 1080p HD video is concerned, you can't cheat the color gamut of REC709, because that is all the monitor can show, whether 8 bit or 10 bit. Dynamic contrast manipulation by the TV does not give you anything that is not already contained in the REC709 encoded video. It may or may not look better than just straight video into a properly calibrated monitor. There is no standard for active contrast manipulation, it is different for every TV manufacturer. You would not want to grade using such a setting, because you don't know what it will look like on any monitor other than your own. This is just marketing hype.
If you look at test reports for most consumer HDTV's you will find that many will not actually reproduce 100% of the REC709 gamut, much less anything broader.

4k REC2020 compliant UHD TV's will show a much broader color gamut, but most of the first gen 4k TV's still only cover the REC709 gamut.
The REC709 HD standard will not change.

I will add that I don't see many samples of video grading that even come close to using the full gamut of REC709.

 

[Samuel H]

* You're right, my statement that "A 4K image with 8-bit 420 color has enough information to give you a 2K image with 10-bit luminance and 8-bit 444 color" is only 100% true if you record uncompressed.
* On my NEX-5N (1920x1080 at 24fps with 8-bit 420 color encoded into 24Mbps), the 420 part throws away half the original information, then the codec throws away 96% of what it receives. Mostly redundant stuff, really...

 

[joe 1008]

* You're right, my statement that "A 4K image with 8-bit 420 color has enough information to give you a 2K image with 10-bit luminance and 8-bit 444 color" is only 100% true if you record uncompressed.
* On my NEX-5N (1920x1080 at 24fps with 8-bit 420 color encoded into 24Mbps), the 420 part throws away half the original information, then the codec throws away 96% of what it receives. Mostly redundant stuff, really...

So it's no real workaround using 4k footage with a low end codec to get better quality HD footage. I was wondering how the 4K 60Mbps of Sony's FDR-AX100 would play out in HD compared to the native HD codec with 50Mbps. Now I suppose it wouldn't be worth the extra work.

 

[Razz16mm]

So it's no real workaround using 4k footage with a low end codec to get better quality HD footage. I was wondering how the 4K 60Mbps of Sony's FDR-AX100 would play out in HD compared to the native HD codec with 50Mbps. Now I suppose it wouldn't be worth the extra work.

Only if the post downscale is better than what the camera has built in. Could be some minor improvement, but the camera is working from 14 bit or 16 bit raw data internally.

 

[harddrive]

I'm somewhat intrigued with the possibility to get kind of a 10bit HD image out of a 4K 8bit image. It might be as simple as converting H264 HD files into PRORES with the extra benefit of using more of the codec's possible colour space.

I'm assuming that what Samuel H was alluding to was that in a simple case, you average 4 4K pixels to get a single 2K pixel. Averaging the pixels should average the noise as well, hence the idea of 10 bit from 8 bit?

I'm not sure it's quite as straightforward as that, and a lot may depend on the noise level of the camera. Perversely, the more noise free the camera, the less that scenario may be the case.

Imagine as an example 4 pixels in a block, three of which have the 10 bit value 0000000011 (3) and one has 0000000100 (4). In an 8 bit system, rounding will give all four the same value - 00000001 (equivalent to 4in the 10 bit system).

Convert to 2K, and the averaged (10 bit) value in the first case will be 13/4=3.25 - which will round to 0000000011 (3).

Do the same averaging derivation for 10 bit for the original 8 bit case, and the result will remain 4 (0000000100)

That's an extreme example, obviously, but it's to try to show the principle that once you've lost integrity by rounding to 8 bit, you can't necessarily retrieve it by the averaging process.

In a system with a higher noise level (higher than the bit depth) then the averaging may well reduce the noise level - but that's a different matter to bit depth improvement. Practically, I don't think you'll win, and I think the most fundamental feature is the size of the sensor chip. Put more photosites onto it, each will be smaller, hence noisier, but an averaging will reduce the noise level again. Put just enough for required resolution - and they will be bigger, quieter - but you can't then do any further trickery. I suspect the results in each case will be similar.

None of that is to say that you don't get any advantage in doing post work at 10bit after 8bit acquisition - you do, by less rounding truncation - but I don't think it's true to equate 4 8-bit
samples with a single 10-bit sample.
The colour subsampling is a different matter. Both 4K 4:2:0 and 2K 4:4:4 have exactly the same number of chroma samples per frame - 1920x1080 for quad-HD. Hence to all intents and purposes then 4K 4:2:0 should correspond in chroma terms to 2K 4:4:4.

Whenever you hear, "4K, but only 4:2:0!?!" then just remember that such will have twice as many chroma samples as 2K 4:2:2 (1920x1080 versus 960x540)

 

[joe 1008]

You make a valid point here demostrating that the result can't be real 10 bit. But won't there be in most cases a result from averaging those four 8bit pixels that could be described with more precision by 10bit?

Or, in a real world situation: Even grading 8 bit carefully easily causes banding. Could a "fake" 10 bit codec protect my footage better from banding? (Of course it would have to be very clean, otherwise those 10 bits, as you wrote, would only show grades of noise - which itself, however, helps to avoid banding.

But ayway, 4:4:4 alone might be worth the conversion to HD, adding the advantage of an easy 2K workflow.

 

[Guest]

You make a valid point here demostrating that the result can't be real 10 bit. But won't there be in most cases a result from averaging those four 8bit pixels that could be described with more precision by 10bit?

Or, in a real world situation: Even grading 8 bit carefully easily causes banding. Could a "fake" 10 bit codec protect my footage better from banding? (Of course it would have to be very clean, otherwise those 10 bits, as you wrote, would only show grades of noise - which itself, however, helps to avoid banding.

But ayway, 4:4:4 alone might be worth the conversion to HD, adding the advantage of an easy 2K workflow.

The ugliest 8bit banding is usually in much larger blocks than 4pixels, so rounding down, and even approximating pixels into a 10bit space won't help 8bit banding issues for the most part. Also, it does not increase color accuracy.

Though, I wonder if there are any plugins or applications which address 8bit gradients by creating intermediate steps?


wall gradient crop by onesielaff, on Flickr


8bit Grad Graded and Compressed by onesielaff, on Flickr

 

[David W. Jones]

Sometimes I have to laugh at this whole it's gotta be 4K or it's crap mentality. My current camera will shoot 5K, and my last camera would only shoot HD. But my HD F35 had a much more pleasing look than my 5K Epic. Replaced an XH-A1 with a Red One M, replaced the Red One M with an AF100, replaced the AF100 with an F35, replaced the F35 with a Red Epic X. No telling what camera we will use next year.

 

[Grug]

Downscaling 4k to HD doesn't give you more colour resolution - it gives you more colour resolution 'per-pixel'.

There's a difference.

 

[Samuel H]

Exactly, you get more color resolution per pixel, but less pixels. It doesn't create any new information.

Let me go through the details because I think there's some confusion as to what I meant with the 10 bit thing...

In 4K 420 footage (let's make it UHD for simplicity), you have 3840x2160 8-bit luminance values, and 1920x1080 8-bit chroma value pairs. It's clear already that you're going to end up with 444 8-bit color because all the info is there to begin with. For luminance, you're mixing four 8-bit values to get the value of the new (bigger) pixel. When you mix four values ranging from 0 to 255, how many options do you have for the result? 1024 different options. That's one 10-bit luminance value, which you get by adding the four 8-bit values, instead of averaging them. You could also average them and stay in 8-bit world, but then you're throwing information away.

Now, if compression is giving you a macroblocked image to begin with, there will be much less variation among those four values, and all this flies out of the window. So, the devil is in the details.

BTW, all you have to do to make good use of this is to drop your 4K file into a 1080p timeline in Premiere and give it 50% size, no transcode necessary.

You can test it right now with your current footage: go for a clip where you wished you had 10-bit color, make it 50% size, do you still see the issues that were bothering you? It will depend on the camera, on my NEX-5N it would probably be a yes, on the C100, even though it is also H.264, I would guess it's a no. On the NEX, the problem is not 8-bit, it is a lousy implementation of the codec.

 

[joe 1008]

You can test it right now with your current footage: go for a clip where you wished you had 10-bit color, make it 50% size, do you still see the issues that were bothering you? It will depend on the camera, on my NEX-5N it would probably be a yes, on the C100, even though it is also H.264, I would guess it's a no. On the NEX, the problem is not 8-bit, it is a lousy implementation of the codec.

Good one.

 

[joe1946]

From what I seen at CES all the the TV companies are putting more DR and a wider color gamut in UHD TVs than HDTVs. Just because some people can't see the difference in a 2mp image from an 8mp image does not mean technology will come to a stop.