Resolution concerns, 2K 4K?

[Deanan]

From my days as a graphics programmer drawing vector graphics, the predominant aliasing effect was getting jaggy edges and anti aliasing ment getting rid of those jaggy edges. That kind of aliasing would definitelly go away when outputing to a higher resolution and then downsizing using averages of 4x4 pixels to a resulting in a single pixel. "

When we talk about aliasing we're talking about the full suite of
aliasing problems you can have. Edge aliasing (jaggies) is one effect
of aliasing. There's also nyquist related aliasing ( moire/sampling patterns caused by fine patterns like certain textiles) and chroma aliasing caused by nyquist issues reacting differently in each color channel (sometimes made worse by inadequate IR filtering).

The last one is the most offensive in motion and consequently downsampling doesn't help because the aliasing has created artificial patterns that spread beyond single pixel variations.


Deanan

 

[Graeme Nattress]

Clearly shows the extra resolution at 4k.

Graeme

 

[Karl H]

oh yeh, clearly there is more resolution, but it doesn't look 4x the amount. My initial dispute was not whether you could tell the difference between 4K/2K as a rule, but whether you could with Red footage going through the same process. I was hoping there was some technical reason why it didn't look as high res as I thought.... and there was (focus, sharpening, exposure, F-stop, 3K real world resolution.... all contributing)...

My concerns have been addressed at least technically speaking

I've yet to see images that match the clairty of the Milk girls footage. I'm hoping it will come in a few days. My guess is the Milk Girls footage was actually lit and therefore nothing was really pushed in post to reveal the noise. Some of the images posted so far are either at high ISO's, shot with low shutter speeds, or have been developed from a few stops under - therefore revealing more noise and compression. Thats not a complaint, it's exactly what I wanted to see. You dont learn a lot from an image exposed with a range of 4 stops due to great lighting.

In the coming weeks I think we'll see the best and worst Red has to offer.

 

[Graeme Nattress]

4k would never have 4x the resolution of 2k though - it would have at most double the linear resolution, and of course, 2k from a 2k sensor should not have 2k resolution either - now that would be the fair comparison - take a Silicon Image 2k shot and a RED 4k shot and see.

Graeme

 

[Karl H]

Im sure we would see a big difference from those comparisons.

AFAIK though this is a single chip CMOS problem only right? If you had a 3CCD system of 2K then you would in theory get a full resolution 2K image from that array?

BTW Graeme, thanks for answering all these question, its much appreciated.....

you never answered my 3K Scaled Raw question though

 

[marco somaini]

You have to compare apples to apples....

You have to compare apples to apples....

we have a lot of people talking about resolution differences. Well, what I have noticed so far, 4K(4520 X 2540) on red is recorded at 72DPI, which is great, but if you compare that to a 11Mpx camera, it records 2731 x 4096 pixels but at 240DPI! that is a lot more information that gets recorded in a still image. So if you really want to compare a canon shot with a red image pulled of the web so for, you need to compare it to an about 1.5Mpx still camera! Just my observation and please red team, let me know and I am wrong on this. I love what yo have done and you guys, keep in mind that 1080 as been the biggest we could record so far on our budget, so cut them some slack, this is not a still camera, it is simply amazing digital video!

 

[Brook Willard]

...or you can just change the 72dpi file to 240dpi... the pixel count doesn't magically change with ppi resolution.

The 4K images posted so far are all 4096x2048.

 

[Eirik Tyrihjel]

Dpi has no meaning in video aquisition - it´s a measure of how detailed a printed or scanned image is - Dots per inch.

I can´t explain it better Marco, I am sure someone else can.

For video the only thing you need to look at is the number of pixels, DPI only has a meaning when you define the size of text in an editing application.

 

[Álex Montoya]

Marco needs an education

 

[Joel Kaye]

it records 2731 x 4096 pixels but at 240DPI! that

Unless I've lost my mind pixels are pixels. DPI is just a representation of how those pixels are going to be place on a device... usually some type of printer. So basically your whole post is wrong.

 

[Jack Wester]

it records 2731 x 4096 pixels but at 240DPI!

DPI or dots per inch is synonymous to pixels per inch. If you show Red footage on a one by one inch display its DPI well be several thousands.

DPI is a good measure on resolution for a output display or on a scanner. Given that the surface has the same size, higher DPI always means more pixels. But in this case it has no meaning. A pixel is a pixel. It can have accuracy, channels and range (bit depth or bit length), but other than that, its a pixel.

No worries though. I've posted more errors on this forum than anyone :whistling:

 

[GlennChan]

AFAIK though this is a single chip CMOS problem only right? If you had a 3CCD system of 2K then you would in theory get a full resolution 2K image from that array?

Not really.

To avoid aliasing, you need to apply optical low pass filtering on your 3 sensor camera. So this means that you won't get full 2K resolution.
Or, you could not apply OLPF. But then you'd have aliasing.

you never answered my 3K Scaled Raw question though

Scaling RAW data would be weird. It probably doesn't work well?

 

[Jack Wester]

The last one is the most offensive in motion and consequently downsampling doesn't help because the aliasing has created artificial patterns that spread beyond single pixel variations.

Got it. Makes sense. Thanks!

 

[Floris Liesker]

Not really.

To avoid aliasing, you need to apply optical low pass filtering on your 3 sensor camera. So this means that you won't get full 2K resolution.

And you'd need a prism to feed the 3 CCD's which would introduce chromatic aberration and reduce your effective resolution as well.

 

[Damien Molineaux]

Im sure we would see a big difference from those comparisons.

AFAIK though this is a single chip CMOS problem only right? If you had a 3CCD system of 2K then you would in theory get a full resolution 2K image from that array?

BTW Graeme, thanks for answering all these question, its much appreciated.....

you never answered my 3K Scaled Raw question though

Actually he did :

Darkline, if only it were so simple.....

The measured resolution from the 4k is over 3k, but say we downsampled it to 3k, we'd now need to be in RGB, not raw, and therefore the file size just got bigger, not smaller. Downsample to 2k, and you've got a 2k full of nearly 2k resolution, but the file size is 3/4 of that of the 4k and you've lost 1k of resolution for that. That's why 4k is best recorded as 4k.

Graeme

You obtain your real resolution once you're in RGB. You can't save a downsampled 4k raw to 3k raw and hope to get a real 3k RGB.

The force of Raw is just that "Raw data", what your sensor sees (ie. records), no downsampling, interpretation, etc.

May the force be with you,
Damien

 

[Karl H]

Yeh I was being a bit cheeky. Jim mentioned a while ago that Graeme had maybe figure out a way to deliver 2K Scaled Raw. I dont really know how it would be achieved as what you said makes sense.

But Jim mentioned it a few times that it was underdevelopment at 2K scaled, not windoowed, so I was prodding and thought, if you can do 2K you should be able to do 3K

 

[Jannard]

We still have some amazing things coming (at no extra charge) for the RED ONE. These cameras are not full featured enabled (I'm sounding like a broken record). Some of the features you have heard about, some you haven't. Owning a RED ONE will be like having Christmas on a regular basis. Just about the time you think you have it down pat... poof! There is another surprise that shows up in your email account.

Jim

 

[Graeme Nattress]

Thanks for your good answers there Glenn. I was flying to Amsterdam so thanks for giving the answers I would have given.

Graeme

 

[Brook Willard]

Owning a RED ONE will be like having Christmas on a regular basis.

Where's the :laugh: icon when you need it.

 

[egon touste]

Yup, that's all you can do for drawing digital imagery - plot at a much higher resolution and average, or better still calculate the percentage of the line passing through a pixel and plot it that brightness.

However because aliasses are frequency mirrored artifacts, they fold back into the data in such a way that they can't be removed by downsampling techniques. For downsampling, basically you low pass filter the data, and then discard any data not needed to create the new downsampled image. So for 4k to 2k, you'd filter with a filter that removes any frequency above that which 2k can sustain, then discard every other sample. Because aliasses fold back into the data, the contaminate the full frequency spectrum, so no filter bases approach can remove them without removing real information also.

Graeme

ha haha graeme!!

..then thats your mistake.. thinking that just because sampling theory seems to demand a low-pass filter on the data first to satisfy niquist etc etc.. wrong approach IMO.. very wrong

there's a long line of scientist who think like this ..and whilst there are good reasons to do this outside of cinematography.. its why ive HAD to do my own in-house downsampling

always take the "longest route" is my advie.. send all 4k pixels -> 2k/1k output etc, floating point, weighted averages.. NEVER THROW AWAY PIXELS (what you would call unresolvable pixels) ..
(so for example, resampling an anamorphic image, say 2048 x 2048 -> 4096x1536.. would require 4096x2048 resamples..taking the longest line from source OR destination)

i used to think it was not possible to experience aliasing affects using 'human eyes'.. but my bathroom curtains proved me wrong..

realising that nature DOESN'T LOWPASS FILTER helped me adjust my thinking..testing many other approaches eventually yielded better results from an artistic point of view, than a strictly technical one (and resolution charts look good too

i have seem terrible tests supposed to demonstrate that downsampling can, or cannot reduce noise floor, correct phasing issues etc .. but the tests are laughable, because there are so many rubbish resampling algorithms out there.. all strictly adhereing to sampling theory, all "real-time" (or as i like to say.."compromised") ..and mostly doing a two pass horizontal, then vertical resample .. and diagonals ?! ..