Uhd to hd. Is there a dynamic range improvement

[Bob Hart]

I have heard in conversation that originating in UHD then converting down to HD for assembly and edit confers an apparent improvement in a HD deliverable in terms of colour and dynamic range, versus originating in HD. I realise that the image will have improved apparent crispness but that was as far as I thought it went.

If anyone can point me in the direction of the correct theory and methods, this would be appreciated.

[Steve Game]

I suppose that it depends on how the downconversion takes place. If there isn't sufficient filtering before subsampling, the artifacts will increase the effective picture disturbance maybe appearing as increased noise. In theory, a sensor with 1/4 of the illuminated area per pixel will have a higher noise floor. If the scene is not very well lit, this might be visually detectable thereby reducing the effective dynamic range.

[Pete Cofrancesco]

Originally Posted by Bob Hart

I have heard in conversation that originating in UHD then converting down to HD for assembly and edit confers an apparent improvement in a HD deliverable in terms of colour and dynamic range, versus originating in HD. I realise that the image will have improved apparent crispness but that was as far as I thought it went.

If anyone can point me in the direction of the correct theory and methods, this would be appreciated.

Based on experience and common sense, down converting from a higher resolution doesn’t add color or increase dynamic range.

You can google “video color bit depth” for a more detailed answer

[Cary Knoop]

Originally Posted by Bob Hart

I have heard in conversation that originating in UHD then converting down to HD for assembly and edit confers an apparent improvement in a HD deliverable in terms of colour and dynamic range, versus originating in HD. I realise that the image will have improved apparent crispness but that was as far as I thought it went.

If anyone can point me in the direction of the correct theory and methods, this would be appreciated.

Converting UHD to HD can increase the bit-depth of the luma and chroma channels and reduce chroma subsampling.

For instance, an 8-bit UHD 4:4:4 video converted to HD becomes a 4:4:4 video with all channels being 10-bit due to oversampling.

An 8-bit UHD 4:2:0 video converted to HD becomes a 4:4:4 video but only the luma channel will be 10-bit, while the chroma channels remain 8-bit.

Dynamic range will not change.

[Pete Cofrancesco]

I remembered a while back there was a thread about the same topic. Someone on the internet put out this idea and it’s just a lot of semantics.

Let’s cut to the chase. If you want to color grade you need a camera capable of filming above 8bit.

[Mark Williams]

Don't know about measurable color and dynamic range improvements but a couple of years ago I shot a lot of 4k with my GH4 and brought into 1080 projects since I didn't have 4k compatible software back then. It looked vastly better than footage shot in 1080.

[Gary Huff]

Originally Posted by Pete Cofrancesco

Let’s cut to the chase. If you want to color grade you need a camera capable of filming above 8bit.

Let's cut to the real chase. No one here is grading in any way that stretches the capabilities of 8-bit color. No, adjusting contrast/saturation and applying LUTs is not grading.

Aside from that, downsizing from 4K to 1080 will give you a better 1080 image on some cameras than the 1080 output, but that's it.

[Mark Watson]

As for getting some added bit depth, I don't believe it. And I've never seen an actual example of it working by anyone who says it is possible. Give me the details of what you did and I should be able to replicate the result.

I have downscaled from UHD plenty times and when I inspect the video properties, there's no magical 4:4:4 or 10 bit from 8 bit footage showing up. I have downscaled from HD to SD via software such as in Sony Vegas, Handbrake, Virtual Dub and TMPGEnc, and also via hardware methods using a special DVD burner that takes HDMI input and will downscale on the fly to burn immediately (as you're shooting) to a DVD and also with a Decimator MD-HX. Never got any of that 4:4:4 or 10 bit from 8 bit goodness.

I bought some 10 bit cameras.

What I DID see with a UHD to FHD downscale, was a nicer image than if I'd shot it in FHD from the same camera. More crisp, like it got a resolution bump. Another thing I found is that when shooting a low-light scene, the grain seemed finer, which made for a better result when running it through NeatVideo.

[Cary Knoop]

Originally Posted by Mark Watson

I have downscaled from UHD plenty times and when I inspect the video properties, there's no magical 4:4:4 or 10 bit from 8 bit footage showing up. I have downscaled from HD to SD via software such as in Sony Vegas, Handbrake, Virtual Dub and TMPGEnc, and also via hardware methods using a special DVD burner that takes HDMI input and will downscale on the fly to burn immediately (as you're shooting) to a DVD and also with a Decimator MD-HX. Never got any of that 4:4:4 or 10 bit from 8 bit goodness.

Just because you have used software that cannot handle this correctly does not mean it can't be done.

The chroma channels are already HD for a UHD clip, that's where the 2 in the 4:2:0 comes from, you only need to resize the luma channel!
So obviously the result will be 4:4:4, without doing anything to the chroma channel.
By having 4 samples in the luma channel reduced to 1 you increase the bit-depth by 2, it's called oversampling!

[Gary Huff]

Then you could clearly show a demonstration of this in action could you not? Oh wait, you can’t.

[Pete Cofrancesco]

Originally Posted by Gary Huff

Let's cut to the real chase. No one here is grading in any way that stretches the capabilities of 8-bit color. No, adjusting contrast/saturation and applying LUTs is not grading.

That’s presumptive to claim to know what everyone does and to speak on their behalf.

If you re read his post he is asking will downresing improve color and dynamic range. Which it doesn’t, filming in 10bit log and grading it, on the other hand, will do both. Whether it’s worth his time and money is up to him.

[Cary Knoop]

Originally Posted by Gary Huff

Then you could clearly show a demonstration of this in action could you not? Oh wait, you can’t.

It's very easy in Vapoursynth:

- open the stream
- apply a conversion filter to convert the bit-depth to float32
- apply a resize filter (resize only the luma plane to HD, the chroma planes are already HD)
- apply a conversion filter to convert the bit-depth to 10 bit
- export the Vapoursynth output stream with FFmpeg to ProRes 444

[Gary Huff]

Originally Posted by Cary Knoop

export the Vapoursynth output stream with FFmpeg to ProRes 444

That's cool. I can do that in any number of programs. What you haven't demonstrated is any benefit whatsoever, and your method does not add anything to a clip that allows it to grade like a truly captured 12-bit 444 clip.

[David Newman]

The effect is minimal, and it relies on noise or gradients that are not quantized away by the source encoder. e.g. 8-bit values in a 2x2 matrix:
[10,12,
13,14]

in 10-bit it contains no more inform:
[40,48,
52,56]

When downscales by 2 in each direction, these valued are averaged to:
49

Which only exists as a 10-bit value, as in 8-bit that would be 12.25.

If the noise is quanizated away:
[10,10,
10,10]

There is not more dynamic range to achieve through scaling.

As all common 8-bit sources are compressed, the scaling is less effective than the theoretically potential gain. From a compression camera you might gain about 1-bit, from every 2x2 downscale. Yes it is subtle. The other benefits of 4K shooting are more significant, but the math shows it is true, that precision preserving downscales can improve dynamic range.

[Gary Huff]

Originally Posted by David Newman

The other benefits of 4K shooting are more significant, but the math shows it is true, that precision preserving downscales can improve dynamic range.

Sure, the math may work out, but the math isn't what matters, the final image is.