idea-can display convey higher resolution than its pixel count

[Piotr]

Strange questions come to my mind today . Inspired by the PsF and V1(precision pixel offset and diamonds) threads I wonder if a 540x960 or 768x1377 flat displays could possibly in our eyes/minds have the same resolution as 1920x1080 displays when showing PsF stream.
If we have a 1080x1920 psf stream over 50Hz, consisting of 50 fields 1920x540 per second how perception works if this is watched on a proper 540x920 display? By proper I mean that each field is displayed 1:1 one after another. Does our eye/brain gain from the fact that each field brings half information of the same frame and puts this frame together so we could have the impression of watching 1920x1080 ?
Would the effect be better with 768x1377 displays?
This came to my mind after reading a comparison of HD ready and Full Hd displays concluding that there is no difference in perception of 1080i signals.
Piotr

[Piotr]

Strange questions come to my mind today . Inspired by the PsF and V1(precision pixel offset and diamonds) threads I wonder if a 540x960 or 768x1377 flat displays could possibly in our eyes/minds have the same resolution as 1920x1080 displays when showing PsF stream.
If we have a 1080x1920 psf stream over 50Hz, consisting of 50 fields 1920x540 per second how perception works if this is watched on a proper 540x920 display? By proper I mean that each field is displayed 1:1 one after another. Does our eye/brain gain from the fact that each field brings half information of the same frame and puts this frame together so we could have the impression of watching 1920x1080 ?
Would the effect be better with 768x1377 displays?
This came to my mind after reading a comparison of HD ready and Full Hd displays concluding that there is no difference in perception of 1080i signals.
Piotr

I already spotted a fault in this reasoning. If we cut horizontal resolution, this information is lost forever. PsF is a trick that works through temporal resolution into vertical space. So the question should be:
Can eg. 768 x1377 (1920?) flat display picture look the same to us as a full 1080x1920 display picture both carrying 1080i/psf 50 hz signal--due to the eye/brain processing ? Or do I need to get some sleep?

[infocus]

This came to my mind after reading a comparison of HD ready and Full Hd displays concluding that there is no difference in perception of 1080i signals.
At the very least, that quote needs a strong qualifier - "in the vertical direction". A full HD display will be 1920 pixels horizontally, an HD ready one less. This assumes the camera was capable of that resolution in the first place.

Theoretically at least, a full HD display doesn't need to deinterlace or scale a 1080i signal, just pixel map. What they do in practice may be another matter

[StevenBagley]

Theoretically at least, a full HD display doesn't need to deinterlace or scale a 1080i signal, just pixel map. What they do in practice may be another matter

Oh it does need to deinterlace -- if you just pixel mapped, you'd be acutely aware of the black lines between the lines of the fields. You have to remember that a CRT's image is made up of a continuously moving tiny dot of light while an LCD, Plasma or SED display is permanently illuminated for each field for 20ms before changing and again being permanently illuminated for the next field's 20ms period.

What you'd end up with is the video image, but superimposed on that would be a series of black lines moving up or down the image, due to the eye being able to track the black lines.

You can't just scale each field up from 540 to 1080 either, otherwise you'll end up with a very nicely filtered 540p panel.

The only option to displaying an interlaced signal properly on an fixed-pixel display is to deinterlace it.

Steven

[infocus]

What you'd end up with is the video image, but superimposed on that would be a series of black lines moving up or down the image, due to the eye being able to track the black lines.
Don't follow that - can you elaborate, Steven? Where do the black lines come from?

Thinking about it I wonder if it's something to do with alternate lines refreshing at different times.........?

Incidentally, for reconstructing p from psf, does the psf signal carry a flag to identify what it is, or does a receiver need to work it out?

[StevenBagley]

Don't follow that - can you elaborate, Steven? Where do the black lines come from?

Thinking about it I wonder if it's something to do with alternate lines refreshing at different times.........?

Yes it is due to lines not being lit by the current field (so for an odd field, all the even lines would be unlit).

If you think about a video signal being a sampled representation of a cube, with the axes, x,y and time then the problem might become slightly clearer if we take a 2D slice of that cube which just contains the y and time axes. In the following graphs, time is represented along the x-axis, and the lines of the odd field (in a 20-line TV system :) ) along the y-axis. The shaded parts represent what is being displayed on screen. So for a CRT:
http://www.eprg.org/tmp/graph-crt.png

Each line of the field is only illuminate for a tiny time period every 40ms, the rest of the time it is black, note also the slope of the line, since each line (and pixel for that matter) is lit up sequentially.

Now, with an LCD:

http://www.eprg.org/tmp/graph-lcd.png

Every line of the field is lit for 20ms, and is then dark for a further 20ms etc.

Now the eye has to integrate these samples into fluid motion, and if I remember my sampling theory right, reconstruction works best with point sources rather than huge square waves (it's why there is so much processing involved in audio DACs, such as adding noise to the signal that is then filtered out).

Steven

[infocus]

Hmmm. I'd sort of assumed that in a plasma display with pixel mapping of an interlaced signal, each pixel illuminated as required, and stayed like that for 40mS, then changing to the value required by the next frame. Same for the pixels of the opposite field, but obviously with a 20mS delay compared to the first field.

If that isn't the case, do you know why it's not done?

[StevenBagley]

Hmmm. I'd sort of assumed that in a plasma display with pixel mapping of an interlaced signal, each pixel illuminated as required, and stayed like that for 40mS, then changing to the value required by the next frame. Same for the pixels of the opposite field, but obviously with a 20mS delay compared to the first field.

If that isn't the case, do you know why it's not done?

Because if you did that it'd end up a) looking filmised due to each field being shown twice and b) having excessive combing visible on anything moving, since you'd be able to see two fields at any point in time.

The way to do it would be to have a strobe backlight and that flashed once for every field, after the pixels had twisted to their new positions, and then flash it. Then switch to displaying field 2 and so on. In fact, I think this is what Philips are doing now in some models (using a scanning backlight rather than a strobe -- see http://www.press.ce.philips.com/apps/c_dir/e3379701.nsf/0/2515E879E57E9131C125705B0032FB00/$File/ClearLCD_Aptura%20technical%20backgrounder.pdf)

Reading that I'm not sure if a Plasma display panel would have the same problem, since they work at much higher refreshes anyway (so that they can do temporal dithering to get the greyscale). I guess an easy way to find out would be to point a video camera at one and twist it by 45o and see if the image looks splits in a similar way to a CRT.

Steven

[Alan Roberts]

I was away yesterday (planning HD training in Bristol), or I'd have dived in much earlier.

All pixel-based displays must de-interlace. Any display that lights up a pixel and leaves it lit for 40 ms will exhibit extreme smear on motion and will be dire to watch. Plasmas don't light for much more than about 5 msec total per 1/50 second, this is partly because the light modulation is temporal rather than by magnitude, so you get a series of flashes with a bse frequency of 50Hz, but with lots of complex higher frequencies because of the sub-field modulation needed to get a decent grey scale from it (remember than a plasma cell can only ever be fully on or fully off).

In much the same way, lcds don't leave pixels lit up for the whole field duration. You'll often see figures quoted for this in the spec (typically 8msec is just about ok, but 6msec is much better, a typical crt lights up for much less than 1msec which is why it's so good at motion portrayal).

Steven's explanation of scanning is right, if you take interlaced fields and use them to control the pixels of their appropriate lines on the display, then the intervening lines must be black because you can't have pixels left lit for a long time (it does no physical harm but the pictures have very poor motion portrayal).

One point to make about the diagonal pixel array in the V1, is that it allows for greater horizontal and vertical resolution than you'd get from the same number of pixels in a normal array, but you also get less diagonal resolution. A normal rectangluar array gives a frequency rersponse that's rectangular (square if the pixel spacings are square), and that gives a resolution in the diagonal directions that's 41.4% greater (square root of 2 is 1.414, 141.4%). All that Canon have done is to rotate the array by 45% so that this 41.4% gain (root 2) is now in the horizontal and vertical directions, so the diagonal resolution is now 70.7% (1/root2 = 0.707). Ideally, we'd like the resolution to be circular because they eye has approximately curcular resolution (although research by Glenn and Gelnn in the US shows that resolution is learned, and that in the artificial society we live in we have a greater preponderance of h and v than of random directions, so babies learn to be more receptive to h and v lines, although I can't see how that works because they're moved around a lot, out of their own control, anyway.....).

Bear in mind that no camera can deliver scenic resolution at greater spatial frequencies than it's sensors, and that attempts to reach that limit produce poor pictures (lots of aliasing, spurious patterning), so cameras are generally set up not to attempt to deal with frequencies above about 80% of the pixel limits. So, a genuine 1920x1080 camera will carry resolution up to about 1600x750, anything above that causes problems. I've measured many cameras over the past decade or so, and the cameras that try to do better than that don't make nice pictures.

Exactly the same's true of displays, except in computers. A computer display generally does not have to deal with motion, so resolution can go all the way up to the pixel limit, but if you then try to make that picture move, you instantly get high level aliasing (rippling edges and jazzy detail). So, for tv, we have to limit the resolution that's fed to it to be significantly below the pixel dimensions. And, that's why it makes sense to go for a display with more pixels than you need, the display won't impose itself on the picture, the picture imposes itself on the display,l which is what we want.

Hope that helps.

[Piotr]

"All pixel-based displays must deinterlace". Does it mean that when a full HD (1920x1080) plasma or lcd get a 1080p signal from a BD it will interlace it first before displaying? AFAIK the new generation of BD is capable of delivering 10080p rather than 1080i/psf.

[StevenBagley]

"All pixel-based displays must deinterlace". Does it mean that when a full HD (1920x1080) plasma or lcd get a 1080p signal from a BD it will interlace it first before displaying? AFAIK the new generation of BD is capable of delivering 10080p rather than 1080i/psf.

If they get a true 1080p signal then a 1080p capable panel will display it as is and hopefully pixel-mapped.

You should also find that the display image is identical to being fed a 1080i signal and letting the panel deinterlace it. Although, this assumes that a) the panel has a decent deinterlacer, and b) that the player isn't filtering the signal for 1080i output -- which it should be to remove vertical frequencies that would cause interline twitter on a CRT display.

Steven

[Piotr]

Don't follow that - can you elaborate, Steven? Where do the black lines come from?

Thinking about it I wonder if it's something to do with alternate lines refreshing at different times.........?

Incidentally, for reconstructing p from psf, does the psf signal carry a flag to identify what it is, or does a receiver need to work it out?

I read that a VC1 codec has such a flag (somebody quoted sect 6.1.13, whatever that means), MPEGs don't.

[Alan Roberts]

Steven's right, if a genuine progressive image is presented, there will be no deinterlacing needed (because it isn't interlaced); de-interlacing is needed only when the incoming signal is interlaced or psf, and, with the flag, de-interlacing of psf won't be needed either. Somed de-interlacers are bright enough to work out, from the image data itself, whether the image is interlaced or not, they don't need the flag (but it never does any harm to have it there).