4 replies to this topic

[evil_josh]

http://nextbigfuture...ts-scarlet.html

4000 lines at 30fps.  Should render a "real" 1080 picture even after Mr Kell has had his say...

Edited by evil_josh, 05 November 2011 - 09:58 PM.

[Owen]

Sorry its not that simple.
All digital source must be low pass filtered to suit the output pixel format, therefore it makes no difference what pixel count you use to sample an image it can NEVER have the visible resolution its pixel count would suggest.
1080p can NEVER have 1080 visible lines of resolution, the result is pretty much as Kell predicted.

[evil_josh]

Owen, on Nov 6 2011, 09:36 PM, said:

Sorry its not that simple.
All digital source must be low pass filtered to suit the output pixel format, therefore it makes no difference what pixel count you use to sample an image it can NEVER have the visible resolution its pixel count would suggest.
1080p can NEVER have 1080 visible lines of resolution, the result is pretty much as Kell predicted.

That's just BS.

Kell deals with the sensor, not the display.  It is a trivial exercise to create an image consisting of 540 lines of one colour alternated with 540 lines of another colour and hey presto, 1080 lines of information!

Now that we have a camera which can deliver over 1080 effective lines of resolution, therefore the display can show all of these lines.

[Owen]

I suggest you go do your homework on digital sampling theory and requirements for real world digital video.

To avoid aliasing and artefacts ALL digital video MUST be low pass filtered to suit the output pixel format, MTF is typically about 30% at spatial frequencies equal to 75% of the pixel resolution and typically 10% or less at spatial frequencies equal to the pixel resolution to ensure that spatial frequencies outside the bandwidth of the system are blocked.
This means that there is effectively no response at 1920x1080 in a 1920x1080 image that has not been computer generated at 1:1

As an example the Sony/Panavision Genesis digital cinema camera has a 12.4 megapixel sensor and down scales to 1920x1080 for output yet its output is as described above.

Capturing with many more pixels then required and downscaling helps maintain MTF at spatial frequencies below the limit imposed by the output pixel format as more in band information can be retained, however response at spatial frequencies equal to the pixel grid should always remain very low and spatial frequencies beyond the limit imposed by the pixel grid must be removed to avoid aliasing and distortion.
Filters cannot have an infinite cut off so in band frequencies are always attenuated in order to make sure out of band frequencies are adequately attenuated, therefore 1920x1080 digital images can NEVER have 1920x1080 visible resolution.
If filtering where not used the picture would look dreadful with interline flicker on high contrast details and moiré.

Edited by Owen, 09 November 2011 - 10:05 PM.

[MLXXX]

evil_josh, on Nov 5 2011, 09:56 PM, said:

http://nextbigfuture...ts-scarlet.html

4000 lines at 30fps.  Should render a "real" 1080 picture even after Mr Kell has had his say...

Provided the high resolution camera image is stored retrieved and displayed using a sufficiently dense sampling grid.

To a first approximation the sampling grid needs to have double the horizontal density, and double the vertical density of the target visible resolution, if the image is to retain a fairly smooth non-digital look, as the camera pans across the scene and small details in the scene alternately align with, straddle, and then align with, elements in the pixel grid. (I can even recall reading a recommendation to use a ratio of 3:1, for a particularly smooth image.)

So if the target video visible resolution is 1920 x 1080, storage and display could utilise 3840 x 2160 pixels, to achieve reasonable smoothness.

I've noticed that FTA broadcasters sometimes "cheat" when broadcasting HD sport using an SD stream of 720x576 pixels. Instead of blurring the image when resampling down to to 720x576 they allow through high enough spatial frequencies from the HD source for moiré patterns to become visible (e.g. in the patterning of the tennis court net) and for visible flickering to appear during pans and zooms (e.g.  line markings on a tennis court).   Such obvious artifacts would be unacceptable for a transfer to Blu-ray disc of a high resolution Hollywood movie. The aim with a 1920x1080 Blu-ray disc is to present a smooth image free of digital aliasing artifacts, even if that comes at the necessary expense of a softer image.

You may possibly already be familiar with these matters, evil_josh, but I think many readers would not be. (To some extent all I am doing is reiterating points Owen has already made.)

A key point is that to display a vertically scrolling test pattern of 540 thin horizontal white lines* on a display panel smoothly, a display consisting of 1080 rows of pixels is inadequate. 2160 rows of display pixels may be sufficient. 2160 rows of display pixels would allow the very high resolution camera to move up or down slightly without creating too obvious a shimmering effect on the display.

________

* That is 540 white lines with black gaps separating them, for a total of 1080 "lines". This may be described as 540 "line pairs", or 1080 "television lines".

Edited by MLXXX, 10 November 2011 - 12:02 AM.