117 replies to this topic

[pneu]

Adam-O, on Jun 4 2006, 12:54 AM, said:

Darklord, do you still say that these displays throw away 50% of the information contained in the video signal?  Ikari, do you still dispute this?

I would say, somewhere between 50% and 100% of the resolution is represented with bob de-interlacing.  But it depends on what resolution you are bobbing to, and which method of bob is used.

I honestly wish I never challenged Darklord on this petty issue.  I don't think we are arguing about facts, we are arguing about how we perceive images and that can be a subjective thing.  But generally speaking, if you show 2 images one after the other at 50hz, the brain will put these 2 images "together", and that is something that can't be denied.

It all comes back to what I said on page 1:

Ikari, on Apr 14 2006, 01:07 AM, said:

If weave or adaptive motion deinterlacing [instead of bob] has the effect of containing more image detail when actually looking at it on the screen then I suppose it would be correct to say that they are in fact more detailed images (even though it technically doesn't contain any more picture information).

I honestly thought the argument would end with that reasonable statement.  But things kind of escalated from there  

I would be happy to resolve this dispute with Darklord.  I really don't want to get in another sniping match though (something I think we are both guilty of).

I'm interested to hear what you think, Adam-O!

[Darklord]

Adam-O said:

This thread is golden... especially the first couple of pages!

Hello Adam. I knew it was only a matter of time before you popped your head in here! It’s an interesting thread isn’t it (at least initially…)

Adam-O said:

I read it all when it was originally happening, a month or so ago, but didn't have time to comment back then. Kind of wish I had

Come on guys? Have you resolved everything now?

I backed out of this thread some time ago, not because I felt I couldn’t have kept going with valid responses and accurate technical information to back up my claims (which I believe I did nearly every post) but rather because it was clear that Ikari was seeing things very differently, and we simply weren’t going to see eye to eye on this issue. I also didn’t like the fact that the thread started to take on a bit of a nasty tone (and I admit I was partly to blame for that).

Adam-O said:

Darklord, do you still say that these displays throw away 50% of the information contained in the video signal?

Call me crazy but I still agree with all the industry experts on this one! . I.e. broadcast/video production engineers, ISF calibrators and companies lire Silicon Optix that design video processors for a living. These people all regularly use the “bob throws away 50% resolution” claim, and it’s completely accurate to do so. That is to say that where film sources are concerned, bob only looks at half of the information available to it (one field – or half a frame) and artificially creates a full frame from it. So yes, I still maintain that bob discards 50% resolution that is available to it, that would otherwise be maintained if weave de-interlacing was used. Ikari – Please don’t even think about refuting this again. Our views are both well documented and I have no desire or time to discuss it further!

I understand that Ikari is arguing that some of that detail is still perceived after bobbing in the following up-scaled frame, however it is not even close to the detail and resolution perceived using weave. I’ve seen this time and time again in real world environments with real world players, scalers and displays. There is simply no disputing that weave results in far more detail (and a far better quality image without line flicker, aliasing or moiré) than bob. I think Ikari actually acknowledges this part though, so I’m not trying to prove any point here. Just stating the facts and summarising my position.

Regardless of any detail that bob may or may not maintain, there is no denying it’s a vastly inferior method of de-interlacing, so discussing its merits (and confusing newbies in the process) is in all honestly a waste of time. Its like arguing that a Datsun 180B isn’t quite as crap a car as most people think.. Well who cares when its still crap?

Adam-O said:

P.S. Darklord, I think your guide was an excellent introduction to de-interlacing and scaling. Well done!

Thanks Adam. Much appreciated. I just hope that the long winded discussions that came after it won’t confuse newbies too much.

I honestly think Ikari would have been much better off raising his questions (given they were such minor gripes relating to an inferior method of de-interlacing) in another thread or via email. All water under the bridge now though, and there are certainly no hard feelings here.

[Adam-O]

OK, my 99 cents

Of course there's no doubt that correctly de-interlaced video looks far better than bob de-interlaced video, and I know that both of you accept that.

The thing that I found interesting was Ikari's challenge to the comment that displays which bob de-interlace are throwing away 50% of the resolution. Of course this comment has been made many times by many people and, in a practical sense, is almost true. But when Ikari challenged it to the letter, I found myself having to agree with him.

These displays are not actually throwing anything away, other than the opportunity to merge the two fields in a manner which is more pleasing to the eye. However, every single pixel that exists in the source is still contributing to the final output, so nothing is being simply discarded.

Sure, it's more difficult for the brain to re-assemble the two fields and glean all of the contained resolution, since these natively progressive displays do not have the phosphorus properties of CRTs. But the information is still there. Scaling alignment issues aside, it would be possible within a controlled environment to actually re-create the original fields and then present the frames in a more pleasing manner, as per what normally happens after "correct" de-interlacing. So this demonstrates that all of the information contained within the source is being displayed. Not one pixel is being discarded.

Of course the result (of bob de-interlacing) looks far inferior. The flicker, for a start, is a big turn off. Also, the brain is being relied upon to re-assemble the two fields. Interestingly though, the brain does infact glean the additional resolution, as the animated GIFs that Ikari provided clearly show (you do have to be looking at them in Firefox to get the speed required to correctly witness this effect). It's not pretty, but it is all there.

So I guess a lot of this really comes down to semantics. One side says that 50% of the information is being thrown away. The other side says that, well, actually nothing is literally being thrown away. If you're talking to the letter, the latter is true. If you are talking in a practical, effective sense, yes, data appears to have been discarded.

I guess the other issue I have is with the figure of 50%. I believe that it is easy to dispute this figure. A lot of people assume that you get double the spatial resolution when using an interlaced format, if it's correctly de-interlaced and displayed. (And, I'm only talking about still scenes here... obviously during movement you definitely do not get any increase in resolution due to interlace). So logically, they say, if you get double by using interlace correctly, then you are losing half (50%) if you use interlace incorrectly (bob de-interlace). I dispute this.

Interlace does not really, and never has, provided double the vertical resolution. In order for interlace to look acceptable, television networks must apply vertical filtering prior to mastering/broadcast in order to avoid/reduce interline flicker. This filtering reduces the vertical resolution dramatically. This is one of the reasons why, if produced correctly, 576p can actually look amazingly good. As a progressive format, no interline flicker-prevention filtering need be applied. So you really are benefiting by 576 true scanlines.

Now you may be tempted to say, OK, that's fine for video material, but what about movie sourced stuff, where correctly de-interlacing requires only a simple -- and lossless -- weave? Well, although the de-interlacing step is certainly easier and does not result in de-interlacing artifacts that come about by having to choose between weave and bob on a per-pixel basis, the vertical resolution problem is all still there. Television networks still have to apply vertical filtering to video that is based upon a movie source. If they did not, fine vertical details would flicker uncontrollably on CRT displays and these viewers would not able to stand watching! So, interline flicker is just as much a problem whether the source is truly interlaced or not. It's because of the interlaced nature of the display device (CRT) -- not the source -- that requires vertical filtering to attend to the problem of interline flicker.

If no one was watching broadcasts in interlaced formats using CRT displays, then the networks could stop applying vertical detail reduction, and all would be fine, since progressive displays do not require this vertical bluring in order to provide a stable image. Obviously this is not the case... there will always be viewers with CRT displays (well, you know what I mean).

This, therefore, is the real difference between a 50i format (showing film-based content) and a 25pSF format. Many people say they are effectively the same because a good de-interlacer can simply weave the 50i format to recreate the 25 discrete frames per second at full vertical resolution. Only problem is, these frames are not at full vertical resolution! Being an interlaced format, they have already had vertical filtering applied to handle the interline flicker problem. Remember that 576i and 1080i are designed to be viewed on a CRT where interline flicker is inherent if this filtering is not applied. Since 25pSF is not broadcast to consumers, the above is always the case.

So, if weaving a progressive source delivered via an interlaced medium does not yield double the resolution, then it follows that by not weaving, you are not losing half of the maximum obtainable resolution. So how much additional vertical resolution does an interlaced format provide? Takashi Fujio of NHK in Japan measured a factor of about 0.6. Ie. for 576i, you really only obtain an effective vertical resolution of just 345 lines -- far short of 576. For 1080i, you effectively only get 648 -- less even than 720p!

The 0.6 figure is probably a little harsh. 0.66667 and 0.7 are also widely accepted figures. It really depends on the precise amount of vertical filtering that has been applied to the interlaced source. The point is though, that it's no where near a doubling/halving.

There's another thing that works against the 50/50 argument, and that is chroma format. All mastered/broadcast video for consumers is in the 4:2:0 chroma format. This means that there is only half as much chroma information -- vertically -- as there is luminance. There's also half as much horizontally, but that's irrelevant to this discussion. If there's half as much chroma information vertically, as there is luminance, then even if you are starting with a full resolution source (no interline flicker-prevention filtering having been performed), and you discard half of the vertical resolution, the result is still better than half because not one single line of chroma has effectively been discarded -- there was already only half! The result will have half the vertical resolution, but with 4:2:2 chroma instead of 4:2:0, and this makes a discernable difference. Therefore the overall result is discernably better than a simple "halving" of vertical resolution because really only the luminance has been halved -- all of the available chroma in the source is still being displayed. I have included this paragraph for completeness only -- it's certainly not the dominant argument against the "50% loss" proposition, and in fact, does not really help bob de-interlacing's position, since there's still only a single line of chroma per each two lines of luminance, on a per-field basis. The over-riding point here though, is to stop assuming that our 1080 line source material really does contain as much true resolution as one could optimally contain within a 1080-line image -- it does not!

So in summary...

Displays which bob de-interlace are not actually throwing anything way in literal terms, though in practical terms they certainly are not making best use of the data to display it in a human-friendly manner, and this does affect the human perception of such content.

The reduction in effective vertical resolution when watching something that has been bobbed rather than correctly de-interlaced, is not 50%, but more like around 20%. Of course flicker is also a problem for bobbed stuff, so this also detracts from such an output, but not in actual resolution terms.

As you both agree, good quality de-interlacing is certainly the preferred way to go, and anyone who actually views the outputs of both scenarios would quickly arrive at this view. However, we have to be careful about using throw away lines (pardon the pun!) such as "displays that bob de-interlace are throwing away half of the vertical resolution", despite the fact we've heard and read this stated by others. Sometimes things are said -- even by experts -- to make a point, even though they do not stand up to technical scrutiny if taken literally, though may have been a perfectly reasonable comment within the intended context.

I am happy to be shown why I am wrong in any of the above, though I am not expecting this. (That probably sounds a little arrogant! I don't mean it that way -- basically what I'm saying is bring on the discussion if you have something good to contribute!).

Cheers,
Adam

P.S. As a final thought, think about single-wheel DLP displays. They display each colour at a separate moment in time, relying on the brain to re-assemble each frame of the image as a full-colour, full-resolution frame. A small percentage of people suffer from the "rainbow" effect, and as such, this is not the most desirable manner in which to present an image. However, all of the data is displayed, and no one would suggest that a single-wheel DLP display is throwing away 2/3rd of the image! The brain is an amazingly adaptive device! Ikari's animated GIFs show that the brain does in fact benefit from the increased resolution, even when using bob, it's just not particularly pleasing. Perhaps, just like the rainbow effect, it's more disturbing to some than others.

P.P.S. As a very final thought, here are just a couple of examples of things regularly said by experts that are incorrect, even though they are acceptable:

1. That NTSC runs at a field rate of 60 Hz. When trying to be correct, it's common to state that it's at 59.94 Hz. Even this is not true. NTSC runs at a field rate of 60000 / 1001 which is ~59.94005994! Depending on the context of the discussion, it may be perfectly reasonable, even for an expert, to use the 60 Hz figure for convenience. That doesn't make it true, however, no matter how decorated the expert!

2. That electric current travels from positive to negative. This is the accepted convention, and was formalised prior to discoveries being made (when valves were invented) that clearly showed that electrons actually travel from negative to positive! The convention remained because it was so entrenched. Does this mean that someone who states that electricity travels from positive and returns to ground is wrong? Well, technically, yes, but we can forgive them for that and we understand what they mean (and if we don't, it really doesn't matter anyway!).

[Darklord]

Excellent post Adam. A very well thought out response. I am very impressed by the points you raised and analogies to other technical scenarios you used.

Adam said:

Of course there's no doubt that correctly de-interlaced video looks far better than bob de-interlaced video, and I know that both of you accept that.

Agreed. Weave 2:2 processing for 25fps/50hz film sources, 3:2 inverse telecine weave for 24fps/60hz material, and per pixel motion adaptive de-interlacing for native interlaced 50i/60i video is how I would sum it up.

Adam said:

These displays are not actually throwing anything away, other than the opportunity to merge the two fields in a manner which is more pleasing to the eye. However, every single pixel that exists in the source is still contributing to the final output, so nothing is being simply discarded.

In theory, that is true. However as raised above I’d argue that:

1.) Much of this original detail is lost through uneven interpolation on fixed pixel displays
2.) Much of this detail is not perceived by the brain the following frame.
3.) How much is perceived would often relate to the digital display technology, and even the person watching.

I will concede that Ikari’s demonstration gifs do preserve some of the detail in the following bobbed frame, but I would argue that it’s not overly significant (certainly not as good as native interlaced, and not even in the same ballpark as a solid weave). I would also argue that due to the line flicker and other interpolation artefacts introduced, any benefits of detail being maintained are quickly destroyed.

There is also another big point being overlooked here, and I strongly believe this is the cause of much confusion in the A/V industry, and perhaps even the cause of a lot of the debate in this thread. I’ll bold this next bit as it’s the key to a lot of what we’ve been discussing: No one agrees on what bob actually means. To some it means oldschool line doubling, to other its means single field interpolation, to others its means field blending, and in some cases it even means drop field interpolation where the even line field is discarded entirely! (And in the last case it’s certainly always a 50% loss of vertical resolution!)

I wont go listing all the links to demonstrate what I’m saying here, but spend 5 minutes on Google and you’ll quickly see that there are many different descriptions of “bob” all of which are widely used and are all technically accurate.

In short I would say that bob in it’s most generalised form means “any method of converting a single field to a full frame, without using detail from surrounding fields”. Obviously this can mean many different implementations, but what they all have in common is that they all look very ordinary, as doubling/interpolation means line flicker, chronic jaggies along all diagonal edges, shimmer, moiré, blur, and a soft picture with poor detail.

Bob de-interlacing, in the form that Ikari, you, and I are discussing (where odd and even fields are interpolated to full frames) can preserve some of the detail from one frame to the next. I will formally go on record as agreeing with this just to make you guys happy . However! At very best I’d say it’s a 20% increase over drop field (where the 2nd field is discarded entirely and a full 50% is lost), and because of the many other artefacts introduced (which Ikari’s gifs nicely illustrate!) it’s a still a horribly inferior method of de-interlacing which should be avoided at all costs.

Put simply every time I see bob used on digital display it completely detracts from the enjoyment of the material I’m watching. Without exception.

Adam said:

So I guess a lot of this really comes down to semantics. One side says that 50% of the information is being thrown away. The other side says that, well, actually nothing is literally being thrown away. If you're talking to the letter, well, the latter is true. If you are talking in a practical, effective sense, yes, data appears to have been discarded.

Well said. That’s a very clever and succinct way of putting it.

Adam said:

A lot of people assume that you get double the spatial resolution when using an interlaced format, if it's correctly de-interlaced and displayed. (And, I'm only talking about still scenes here... obviously during movement you definitely do not get any increase in resolution due to interlace). So logically, they say, if you get double by using interlace correctly, then you are losing half (50%) if you use interlace incorrectly (bob de-interlace). I dispute this.

While I know you weren’t specifically referring to me here, just to be clear I’d like to say that I have never claimed the above. I was only talking bob vs weaved progressive. Native interlaced, while better in my opinion than bobbed progressive, still suffers from all the inherent interlaced artefacts that we all know and hate. Even on a CRT capable of fully resolving 1080 lines, you’d never get close to the full 1080 lines on anything other than a still scene, and even then there are issues like Kell factor to take into account.

It’s also worth noting that I’m a huge advocate of native progressive scan in general. I have long argued the benefits of native progressive formats like 720/50p, over filtered interlaced formats like 1080/50i. Particularly in the case of native high frame/field rate video for footage like sport, where there is simply no contest with solidarity, temporal resolution and consistent spatial resolution, between native interlaced and native progressive video.

As far as native 60hz video is concerned, 1080i can accurately be described as 1440x540/60p, with the addition of aliasing, shimmer, line flicker and interlaced blur. It simply doesn’t compare to native progressive 1280x720/60p with a rock solid full 720 line picture at all times.

Just wanted to make that clear. A fan of interlaced scan I am not!

Adam said:

Interlace does not really, and never has, provided double the vertical resolution. In order for interlace to look acceptable, television networks must apply vertical filtering prior to mastering/broadcast in order to avoid/reduce interline flicker. This filtering reduces the vertical resolution dramatically. This is one of the reasons why, if produced correctly, 576p can actually look amazingly good. As a progressive format, no interline flicker-prevention filtering need be applied. So you really are benefiting by 576 true scanlines.

Exactly. I can’t tall you how relieving it is to hear someone else here raise this often overlooked point. Filtering can and does destroy detail on all interlaced broadcasts. I am in complete agreement and have argued this point for years. I remember being shot down in flames when I had the cheek to suggest that native 720x576/50p (if it could be captured, preserved and broadcast in this format) could look dramatically superior to 720x576i for sporting events on Seven HD. Twice the spatial resolution per motion update, no interlaced filtering, far better clarity under motion, and no aliasing, line flicker, shimmer or jaggies. If only material could be captured in this native 50p format I believe it would look very good indeed.

Adam said:

Now you may be tempted to say, OK, that's fine for video material, but what about movie sourced stuff, where correctly de-interlacing requires only a simple -- and lossless -- weave? Well, although the de-interlacing step is certainly easier and does not result in de-interlacing artifacts that come about by having to choose between weave and bob on a per-pixel basis, the vertical resolution problem is all still there. Television networks still have to apply vertical filtering to video that is based upon a movie source. If they did not, fine vertical details would flicker uncontrollably on CRT displays and these viewers would not able to stand watching! So, interline flicker is just as much a problem whether the source is truly interlaced or not. It's because of the interlaced nature of the display device (CRT) -- not the source -- that requires vertical filtering to attend to the problem of interline flicker.

Again, no argument from me. I’ve never been under any misconception that weave somehow returns the video to a state of full resolution. I agree it absolutely does not. However it does return the material to a  state of maximum resolution available from the source device. Bob most certainly does not!

Adam said:

So, if weaving a progressive source delivered via an interlaced medium does not obtain double the resolution, then it follows that by not weaving, you are not losing half of the maximum obtainable resolution.

Ok, but I would argue this is false logic. When people say you are “losing half the resolution” they are talking about half of the resolution that the source device is delivering. (i.e a 1080 line filtered frame split into two halves). No one with any technical knowledge of the process would suggest that you’re getting anything close to the full 1080 lines of the original studio master. MPEG2 compression (and chroma compression which you mentioned above) destroys much of that (particularly in Australia with our woefully over-compressed MPEG2 broadcasts).

Adam said:

There's another thing that works against the 50/50 argument, and that is chroma format. All mastered/broadcast video for consumers is in the 4:2:0 chroma format. This means that there is only half as much chroma information -- vertically -- as there is luminance. There's also half as much horizontally, but that's irrelevant to this discussion. If there's half as much chroma information vertically, as there is luminance, then even if you are starting with a full resolution source (no interline flicker-prevention filtering having been performed), and you discard half of the vertical resolution, the result is still better than half because not one single line of chroma has effectively been discarded -- there was already only half! The result will have half the vertical resolution, but with 4:2:2 chroma instead of 4:2:0, and this makes a discernable difference. Therefore the overall result is discernably better than a simple "halving" of vertical resolution. I have included this!

100% correct. However again, this has nothing to do with the resolution lost through poor de-interlacing. De-interlacing should be considered as one of the last critical steps in the video processing chain. No matter what compression has already been applied (whether it be colour compression, interlacing, MPEG – which are all methods of image compression) poor de-interlacing can further discard resolution (up to 50% depending what method is applied) to an already compressed picture!

Adam said:

As you both agree, good quality de-interlacing is certainly the preferred way to go, and anyone who actually views the outputs of both scenarios would quickly arrive at this view. However, we have to be careful about using throw away lines (pardon the pun!) such as "displays that bob de-interlace are throwing away half of the vertical resolution", despite that we've heard and read this stated by others. Sometimes things are said -- even by experts -- to make a point, even though they do not stand up to technical scrutiny if taken literally, though may have been a perfectly reasonable comment within the intended context.

I see where you are coming from, but I still think that claiming that bob throws away up to 50% resolution is an accurate description (the ‘up to’ is probably the important bit to include). This is of course dependant on what method of bob is employed, the interpolation/scaling that takes place, the resolution and type of digital display (effecting perception of detail in the succeeding interpolated frame) as well as the degree of other artefacts (aliasing, flicker, shimmer, moiré, blur etc) that detract from any maintained detail.

It’s certainly a very grey area, and I think we can now all agree that there is no “right and wrong” when it comes to a very complex area of video processing. In reality the many variables involved suggest that the results will be very different depending on the source, method of de-interlacing, display and perception of the individual.

What I do know, and what isn’t under debate, is that weave looks vastly superior in every case, and that it’s certainly the preferred option for film sourced material.

When you think about it weave can actually be thought of as a method of lossless compression (forgetting any other methods of compression already applied to the picture before interlacing). It’s simply a method of delivery for a format like 1080/25p which can be delivered as 1080/50i, then reassembled into 1080/25p at the other end. Sure it’s not as good as unfiltered native 1080/25p, but its’ certainly a lot better than 540 line fields interpolated to 1080/50p.

[ckent]

Adam-O, on Jun 5 2006, 06:36 PM, said:

Many people say they are effectively the same because a good de-interlacer can simply weave the 50i format to recreate the 25 discrete frames per second at full vertical resolution. Only problem is, these frames are not at full vertical resolution! Being an interlaced format, they have already had vertical filtering applied to handle the interline flicker problem.

Do you have any proof of that? I'd offer lots of examples of SDTV that have graphics on-screen that have alternate lines at full sharpness.

Or are you saying that I will only find examples in native SDTV content, and not in film-derived or HDTV-derived content? In which case, this filter is only applied to content that was higher in definition to start with, yet doesn't apply to native SDTV, which both proves it to be unnecessary as well as showing up some hypocrisy going on there.

______

My only other contribution to the whole (interesting) debate, which I haven't seen mentioned already after a quick scan of this thread, is the overall frame rate.

If I view Ikari's cool animated GIF at a really high frame rate, like 100fps or 150fps or 200fps, then the flicker is entirely gone and you are just left with the question of how much resolution is left. (To represent this faithfully on a 50Hz TV, you need to come to a decision about how it is visualised, and post-process it into a 50p image. This will most likely end with a frame-blending result).

If you think of the "hypothetical extreme" example, oft-quoted, of a screenful of alternating black/white lines, then you will just end up with grey. Fair enough. That means you have just 288p out of 576i.

Also if you change the framerate to extremely slow, say something ridiculous like 0.1fps, you'll end up with 288p as well.

It's only somewhere in the middle, like 25Hz (measuring the full repeat) that the vision is tricked into seeing the detail together.

______

OK. So what else is new? There's one more idea: If you do the bobbing, like line doubling, but introduce BLACK lines instead of doubling the lines, then first of all you'll end up with something a little like interlacing as seen drawn onto a CRT.

But then consider turning up the animated GIF to 200fps (or using a super-duper CRT that can display it). You will end up with FULL resolution!

So this still fits (one!) of the definitions of "bobbing", yet retains full resolution.

FYI, a well-calibrated 100Hz TV should work exactly like this.
Unfortunately, the practice rarely meets the theory.

CK.

[pneu]

ckent, on Jun 14 2006, 02:24 PM, said:

I'd offer lots of examples of SDTV that have graphics on-screen that have alternate lines at full sharpness.

I've noticed that as well - lines in the same frame (from broadcasted SDTV) appear to be unfiltered for some graphics (they lie exactly on a line) but are filtered for other graphics (edges have smoothed bits around the top/bottom of the line).

ckent, on Jun 14 2006, 02:24 PM, said:

My only other contribution to the whole (interesting) debate, which I haven't seen mentioned already after a quick scan of this thread, is the overall frame rate.

Yes, framerate will affect how well bob works.  The aniGIFs I've done are not fully representative of bob because they won't be synch'ed with your monitors refresh rate in the browser, and don't achieve the correct framerate (there is significant tearing).  If it's all synch'ed at 50hz+ like on plasmas, the image will appear to be more solid.  Even higher framerates increase this effect, as you have said.

ckent, on Jun 14 2006, 02:24 PM, said:

There's one more idea: If you do the bobbing, like line doubling, but introduce BLACK lines instead of doubling the lines, then first of all you'll end up with something a little like interlacing as seen drawn onto a CRT.

I was thinking the exact same thing: bobbing with interpolated bits replaced with black lines

note: the aniGIF doesn't seem to work very well unsynch'ed on an LCD screen.

ckent, on Jun 14 2006, 02:24 PM, said:

Unfortunately, the practice rarely meets the theory.

Fair point.  While there are ways to implement bob de-interlacing which preserve more of the resolution, a lot of the time they might not be used and the end result will look poor anyway.

[Stephen Dawson]

Quote

Since the introduction of new digital displays, we’ve run into a problem. The problem being that all digital displays are inherently "progressive". In other words they generally work by illuminating all pixels on screen at once, and therefore can’t operate in interlaced mode like older CRT technology.

This creates a problem with all interlaced formats that must be addressed. The interlaced material must be “de-interlaced” (converted to progressive scan) in order for it to be displayed properly on the digital display. Confused yet? Well hang on as things are about to get a lot messier…..

Actually, this isn't quite correct. Every individual pixel on a panel or LCD/LCoS/DLP projector (all of which I'll henceforth call 'digital displays') is addressable. There is no technical reason why, if the incoming signal matches the display's native resolution, each line cannot be sent to the display panel in the same order in which it was received. Indeed, when I was reviewing some LCD TVs a couple of years ago, I was appalled to find that a couple of models seemed to do just this, resulting in noticable interline flicker!

The reason that deinterlacing becomes necessary for digital displays is because they look like crap otherwise. They look bad because they don't suffer the deficiencies of older CRT TVs. These were fuzzy, so you could get away with a lot. Indeed, not every bit of the theoretical resolution of the signal was realised on these screens because the electron beam would miss the respective phosphor dot a significant proportion of the time. The 'Kell Factor' is used to specify this, and it is typically around 70%: that is, only around 70% of the resolution delivered to the tube would actually be made available to the viewer.

With digital displays (assuming they are not the so-called SD models, with a resolution of 852 by 480 pixels) every bit of the incoming signal's resolution is available to the viewer. Furthermore, because these displays are much sharper, the combing of video-sourced interlaced signals is extremely obvious. This is made even more obvious by the on/off characteristics of a digital display.

A pixel on a CRT display is illuminated when the electron beam hits it. It will not be hit again for another 1/25th of a second. Its peak illumination is achieved very rapidly, and then it tapers off. By the time the dot immediately below it is illuminated (1/50th of a second later), it has already faded significantly, reducing the visibility of combing. With (competently designed) digital displays, two whole fields worth of signal is loaded into a frame buffer (after having been scaled appropriately) and then the entire panel is switched on, with the pixels matching the frame buffer's contents. This is maintained until an instant before the next frame buffer is fully loaded, ready to drive the display. This means that combing lines are shown at full contrast next to each other, making them even more obvious.

Finally, digital displays are typically bigger than CRTs. The larger the display, the more obvious is the combing. Deinterlacing becomes very important once the screen size reaches 106cm, and a projector picture (mine is 212cm) is virtually unwatchable with video sourced material without good deinterlacing.

[Stephen Dawson]

DavoNogo, on Apr 15 2006, 09:04 PM, said:

... can only be done on Film sources that have been converted to an Interlaced signal properly. An example of a bad conversion would be Star Wars Episode II on Ten, which for some reason went through an odd Interlacing process which turned a Progressive film into Interlaced video.

This is a peculiar thing that crops up in the oddest places from time to time. For example, the original DVD release of Independence Day, plus Mad Max and Pulp Fiction all suffer (partially in the first case, totally for the other two) from fields from adjacent frames being packed into the same frame, so that every single frame is badly interlaced. That these are watchable at all is a testament to the remarkable quality of much modern consumer electronic video equipment. See my article Mismatching fields on a PAL DVD ... or how tolerant are our eyes anyway? for details and screenshots.

[Stephen Dawson]

One last comment on this thread. I have to say that the major contributors have made this the place to go for anyone who wants to fully understand issues relating to deinterlacing. It has certainly made me rethink a few of my assumptions. Well done chaps.

[Darklord]

Hello Stephen. Thankyou for your contribution. I take it you are Stephen Dawson from Sound & Image magazine? If so I have really enjoyed your display articles over the years and find myself in agreement with your editorials more often than not. I also check out your blog from time to time which is a great read. I do have a couple of issues with your progressive PAL article from a few years back, but I wont go into that now

Your comments above regarding interlaced scan on digital displays is quite right, and in fact as you probably know ALIS plasmas (the 1024x1024 or 1024x1080 variety) already use a technique similar to native interlaced scan. The difference being they paint all the odd or even lines at once rather than one by one as seen on a CRT. Being a plasma they also don’t have the same degree of phosphor persistence, meaning the illusion of the odd/even lines being onscreen at once isn’t as great. However the technology is quite effective and is a good “cheat” to 1024 lines appearing to beonscreen at once when in reality there are only 512.

The reason I mentioned that digital displays were inherently progressive was solely for the sake of simplicity, as 99% of them are progressive (whether they have to be this way or not I considered besides the point). Unfortunately given the recent debates that have escalated in this thread, my aim of keeping it simple seems to have gone out the window

[ckent]

Stephen Dawson, on Jun 18 2006, 08:18 PM, said:

This is a peculiar thing that crops up in the oddest places from time to time. For example, the original DVD release of Independence Day, plus Mad Max and Pulp Fiction all suffer (partially in the first case, totally for the other two) from fields from adjacent frames being packed into the same frame, so that every single frame is badly interlaced. That these are watchable at all is a testament to the remarkable quality of much modern consumer electronic video equipment. See my article Mismatching fields on a PAL DVD ... or how tolerant are our eyes anyway? for details and screenshots.

The Star Wars episode that went to air on Ten in native HD was actually played out using 3:2 pulldown 1080i60, directly converted to 1080i50 with little or no motion compensation. This totally explained the appearance of fields being weaved, and the complete loss of field dominance. I worked out that in theory it was probably possible (with some manual priming) to recreate the original 24fps from this strange broadcast, give or take 4:2:0 colour issues, and any side-effects from 23.94fps framerate if that applied.

As for plasmas and LCD framerates, I notice that most plasmas don't seem very happy to display motion of any higher than 25fps or so, not that they don't try. You will still see fluidity, but the quality of the colour seems to deteriorate on most panels (some improvements seem to be happening in the last 12 months, in some brands, but nothing utterly successful; hence all the low-motion demo loops in department stores). In the end, if you want to show a line flickering from 0% to 100% levels at 25Hz (ie, 1/50th of a second from black-to-white), it might not actually look like that on the panel's output. For one thing, check out the rapid flickering you're meant to see on the Seven HD loop (make sure it's plugged in via 576p, 720p or some cases 1080i depending on the post-processing; 576i will get flicker filtered quite readily these days) -- the flickering just doesn't appear on plasmas, or if it does, won't be half as noticeable. Which was a good thing, I thought, until I realised sports won't be much good then.

CK.

[Stephen Dawson]

Darklord, on Jun 18 2006, 10:56 PM, said:

Hello Stephen. Thankyou for your contribution. I take it you are Stephen Dawson from Sound & Image magazine? If so I have really enjoyed your display articles over the years and find myself in agreement with your editorials more often than not. I also check out your blog from time to time which is a great read. I do have a couple of issues with your progressive PAL article from a few years back, but I wont go into that now

Yes, that's me. Please read the Progressive PAL article in the context for which is written (I probably should make it clear in the article, as times have now changed.) What I was interested in was DVD players that convert the 576i native signal from the DVD to 576p. That, of course, is quite a different matter to the broadcast of 576p via digital TV since, at least potentially, these signals may have never been in interlaced format (of course, the majority of 576p broadcast material is merely upconverted from 576i).

Since originally writing the thing I've been noticing different things, touched upon briefly in the August 2005 update appended to the piece. I still haven't been able to come down to a final position, though. It is often the case that horizontal resolution is improved a little with a progressive scan output from a DVD player compared to interlaced, and I have put this down to anti-aliasing filtering applied at the DVD player outputs, or at the display inputs, or both, for 576i signals, but not for 576p ones. But, oddly, I have stumbled across the odd display that seems to give better horizontal resolution with a 576i signal than a 576p one! Now that's confusing. Possibly the display electronics were optimised for 576i, with reduced filtering. But since 576i comes in at 13.5MHz, and 576p at 27MHz, the use of the same filter regardless of the input would chop off 576p resolution.

Anyway, to some extent this is becoming irrelevant. With HDMI/DVI signal delivery, the DVD player need not (and should not) filter the output, and any filtering applied at the display end should be performed by the scaler with the sole purpose of optimising picture quality.

But, equally with HDMI/DVI, deinterlacing technology in DVD players takes on much more importance because very few (if any) such DVD players have an option to deliver 576i via HDMI/DVI. They will be converting to (at least) 576p. Of course, since most DVDs are film-sourced, a simple weave is all it takes. But with video sourced material, the quality of the deinterlacer is very important. That's something I'm going to have to pay more attention to in the future with DVD player reviews (not that we do many anymore since they have largely become a commodity item).

Darklord, on Jun 18 2006, 10:56 PM, said:

The reason I mentioned that digital displays were inherently progressive was solely for the sake of simplicity, as 99% of them are progressive (whether they have to be this way or not I considered besides the point). Unfortunately given the recent debates that have escalated in this thread, my aim of keeping it simple seems to have gone out the window

Personally, I'm thankful. There's plenty of simplified material about most technical things out there, readily accessible thanks to Google. It is in drilling down to the satisfying detail that difficulties often arise. As it happens, this thread, thanks to both your and Ikari's dogged defences of your positions, has almost assumed the form of Socratic dialogue, with considerable benefit to all us readers. Adam-O's summary and discussion acted as both excellent mediation and a great roundup of the conclusions.

Stephen Dawson, on Jun 18 2006, 07:15 PM, said:

Indeed, when I was reviewing some LCD TVs a couple of years ago, I was appalled to find that a couple of models seemed to do just this, resulting in noticable interline flicker!

On thinking further about this, it strikes me that the flicker is more likely to have been due to an inappropriately applied 'bob' deinterlacing strategy rather than a 'weave'.

[Darklord]

Stephen Dawson said:

Yes, that's me. Please read the http://www.hifi-writ...an/progscan.htm in the context for which is written (I probably should make it clear in the article, as times have now changed.) What I was interested in was DVD players that convert the 576i native signal from the DVD to 576p. That, of course, is quite a different matter to the broadcast of 576p via digital TV since, at least potentially, these signals may have never been in interlaced format (of course, the majority of 576p broadcast material is merely upconverted from 576i).

That’s fair enough Stephen. I remember reading the article when it first appeared in Sound & Image and overall I thought the points you raised at the time did have some very real merit. Particularly as back then CRTs were still the common display device, and as you quite rightly pointed out 100hz 576i can provide an experience close in visible resolution to 576/25p (in the case of 24/25p derived film that is – native 50i/50p video is a different matter). However even at the time I believe that you did overlook factors such as aliasing, line flicker and moiré, all of which are typically eliminated by weave de-interlacing, resorting the fields to full solid 576p frames. These artefacts are reduced, but still present on 100hz interlaced displays. From my experience weave de-interlaced 576p also tends to have a more solid “film-like” picture than content presented as 576i @ 100hz.

As you know things are very different now with digital displays being the most common form of display for home theatre enthusiasts, most of which require material to be de-interlaced for display. As the deinterlacing article I wrote attempted to pint out, the methods and quality of deinterlacing used are an integral part of the final picture quality.

Then of course there is native 576/50p. We very rarely see anything that could be classed as native 576/50p from Seven (and as you know ABC and SBS simply up-convert from 576i), as Seven typically down-convert from 1080/50i (for in studio based material like Sunrise) or 1080/25psf for a lot of their lifestyle shows and movies (Home & Away, The Great Outdoors etc).

When Seven do down-convert from 1080/50i (i.e. Sunrise) to 576/50p, the material has a full 540 lines (up-scaled to 576p) per 50th of a second vs 288 lines per 50th of a second of the SD feed. This does result in more visible spatial resolution with the same fluid 50hz motion update of the SD 576i source. Unfortunately because it’s derived from an interlaced source (one 540 line interlaced field from the 1080i source becomes a full 576p frame) there is still line flicker, and detail “bobs up and down”. No were near as good as native 720x576/50p would look, that’s for sure. Still, it’s a significant improvement over 576/50i.

Film based 576p broadcasts are of course different, as aside from lack of filtering (which reduces horizontal resolution of the 576i boradcast) they theoretically contain no more data than the 576/50i broadcast (as that can be weaved back to full 576/25p).

Stephen Dawson said:

Since originally writing the thing I've been noticing different things, touched upon briefly in the August 2005 update appended to the piece. I still haven't been able to come down to a final position, though. It is often the case that horizontal resolution is improved a little with a progressive scan output from a DVD player compared to interlaced, and I have put this down to anti-aliasing filtering applied at the DVD player outputs

This may very well be the case. Filtering of interlaced content is used by TV networks for the majority of TV broadcasts that go to air, to cut down on the amount of flickering and aliasing on CRT displays. It’s quite possible (and more than likely) that DVD players also apply a horizontal filter to interlaced outputs.

Stephen Dawson said:

or at the display inputs, or both, for 576i signals, but not for 576p ones. But, oddly, I have stumbled across the odd display that seems to give better horizontal resolution with a 576i signal than a 576p one! Now that's confusing. Possibly the display electronics were optimised for 576i, with reduced filtering. But since 576i comes in at 13.5MHz, and 576p at 27MHz, the use of the same filter regardless of the input would chop off 576p resolution.

That’s also possible, but this anomaly may also come back to de-interlacing (at least in part). Often the quality of the deinterlacing in a good quality digital display will be superior to that in cheaper progressive DVD players. Pioneer’s latest plasmas for instance have very good motion adaptive SD de-interlacing that can outperform the de-interlacing in a lot of budget progressive DVD players. I’ve sometimes advised people to leave their player set to interlaced output and let the display do the deinterlacing as it can provide a better picture. However with most displays the reverse is nearly always true, and a progressive scan DVD player with advanced cadence detection (and diagonal filtering/correction like Faroudja’s DCDi) can be considered an essential purchase to get the best possible picture from DVDs on digital displays.

Stephen Dawson said:

Anyway, to some extent this is becoming irrelevant. With HDMI/DVI signal delivery, the DVD player need not (and should not) filter the output, and any filtering applied at the display end should be performed by the scaler with the sole purpose of optimising picture quality.

Exactly.

Stephen Dawson said:

But, equally with HDMI/DVI, deinterlacing technology in DVD players takes on much more importance because very few (if any) such DVD players have an option to deliver 576i via HDMI/DVI.

That’s right, and it’s of major concern to those with high end external scalers. Most of the time these people are limited to 480i/576i via component as very few player deliver 480i/576i via HDMI. Thankfully some of the higher end DVD players are now starting to provide the option for unprocessed 480i/576i via HDMI.

Stehpen Dawson said:

Of course, since most DVDs are film-sourced, a simple weave is all it takes. But with video sourced material, the quality of the deinterlacer is very important. That's something I'm going to have to pay more attention to in the future with DVD player reviews (not that we do many anymore since they have largely become a commodity item).

I agree. It’s probably THE most critical area when assessing the performance of today’s progressive/up-scaling DVD players.

I’m not sure if you already have it or not, but I can thoroughly recommend getting a gold of Silicon Optix’s acclaimed HQV benchmark DVD. It provides a very wide range of source material ranging from properly flagged 2:2/3:2 material, to common un-flagged 2:2 video, all the way up the bizarre rare cadences (like 3:2:2:3) used by material such as Japanese video sourced anime. This DVD has become the industry standard for evaluating a DVD player’s cadence detection/de-interlacing/noise reduction/jaggie reduction etc. It’s especially handy for reviewers like yourself, as each test is given a score from 1-10, then you add up the results for a final score at the end. It really is the first industry standard benchmark tool for proper evaluation of any device with inbuilt deinterlacing (it can naturally also be used to test the deinterlacing of displays if you leave the DVD player set to interlaced output). For an example of how it is used see this review of the popular Panasonic S-97 DVD Player at Audioholics(they now use this DVD for all player/display/video processor tests).

Stephen Dawson said:

There's plenty of simplified material about most technical things out there, readily accessible thanks to Google. It is in drilling down to the satisfying detail that difficulties often arise. As it happens, this thread, thanks to both your and Ikari's dogged defences of your positions, has almost assumed the form of Socratic dialogue, with considerable benefit to all us readers. Adam-O's summary and discussion acted as both excellent mediation and a great roundup of the conclusions.

Thankyou for your feedback. Nice to know our debating/arguing/bickering/geeky discussions (however you want to refer to them) weren’t lost on somebody

Stephen Dawson said:

Stephen Dawson said:

QUOTE(Stephen Dawson @ Jun 18 2006, 07:15 PM) <{POST_SNAPBACK}> Indeed, when I was reviewing some LCD TVs a couple of years ago, I was appalled to find that a couple of models seemed to do just this, resulting in noticable interline flicker!

On thinking further about this, it strikes me that the flicker is more likely to have been due to an inappropriately applied 'bob' deinterlacing strategy rather than a 'weave'.

Almost certainly a result of bob flicker. Incorrect weaving (i.e. weaving native interlaced content when the fields don’t line up) typically results in “combing” or the “mouse teeth” effect, rather than any introduced line flicker. It can also result in an unnatural halving of the original frame rate as two fields are being display onscreen once, rather than successively (native 50i becomes 25p). Thankfully this is very rare as the vast majority of displays fall back on bob deinterlacing, even when they should be using weave! (As bob is the cheapest, easiest solution method and works for everything).

[ckent]

Darklord, on Jun 21 2006, 04:25 PM, said:

Filtering of interlaced content is used by TV networks for the majority of TV broadcasts that go to air, to cut down on the amount of flickering and aliasing on CRT displays.

Can you back this up? I will claim that any filtering or softening you see, is done on a case-by-case basis by whoever is creating the graphics or film material or video material. There appears to be no filtering done by the "TV network" at the final stage of transmission, therefore, hard edges can and will happen.

I've seen it quite often on news graphics, weather graphics and other graphics, and also is easily visible on the World Cup footage which is shot at 1080i, where you can see the white lines against the green background.

As a result, some 576i footage is directly equivalent to shooting it at 576p then simply removing half the pixels over time.

Equivalently, it is also possible to see film footage on 1080i50 broadcasts that has the same resolution as 1080p25.

CK.

[Darklord]

ckent, on Jun 22 2006, 01:18 PM, said:

Can you back this up? I will claim that any filtering or softening you see, is done on a case-by-case basis by whoever is creating the graphics or film material or video material. There appears to be no filtering done by the "TV network" at the final stage of transmission, therefore, hard edges can and will happen.

You may well be right in some cases CK. I admit I was being very vague with my reference to “TV networks” and you’re quite right when you say that filtering may often be a process that takes place during production.

However, while I don’t have direct evidence to support this, I am still confident in the claim that filtering must often be applied by the networks themselves. The main reason being that they often receive original 1080/24p master tapes from the US, which must be converted to 1080/50i and 576/50i for broadcast. I would be surprised if either:

1.) None of this content is ever filtered – this would result in horrendous flicker on all interlaced displays

OR:

2.) All of this content was pre-filtered during the mastering process -  If so this would mean any worldwide networks opting to use the 720p or 576p standards would be broadcasting material that is unnecessarily filtered and therefore missing valuable resolution.

The only logical conclusion is that filtering is being applied by the TV networks themselves. It may be as simple as option in a Terranex video processor.

This is further backed up by screenshot comparisons of 24p/25p film based material (video based is another matter) between Seven SD/HD. For 24/25p content, weave de-interlaced 576i and native 576p broadcasts should theoretically look very similar (bit-rate differences aside) if filtering was not applied to the 576i broadcast. However its clear that when looking at screenshots comparisons between some shows (Lost being a prime example) that the progressive broadcast has clearly better resolution. This can actually be quite dramatic. I would say in the order of 10-20%.

Unless there is some other factor affecting the quality of Seven’s SD broadcasts, then I cant imagine where this big difference in quality is coming from.

There used to be a very good page dedicated to the topic of interlaced filtering at www.progressivescan.co.uk. Unfortunately it seems the site has been removed.

[ckent]

Darklord, on Jun 23 2006, 12:41 AM, said:

You may well be right in some cases CK. I admit I was being very vague with my reference to “TV networks” and you’re quite right when you say that filtering may often be a process that takes place during production.

It seems to me we're basically agreeing on most of this, then  

A large amount of SDTV =will= show the effects of your filtering, and indeed if you account for every piece of text digitally superimposed on screen, it is impossible to find any that is not antialiased these days. Antialiasing is always going to give soft edges on purpose.

More to the point, if you compare modern examples to older ones, you'll quickly find that in those days everything was softer and less precise anyway, so without the digital manipulation back then it was impossible to find anything as sharp as you could see nowadays.

Quote

However, while I don’t have direct evidence to support this, I am still confident in the claim that filtering must often be applied by the networks themselves. The main reason being that they often receive original 1080/24p master tapes from the US, which must be converted to 1080/50i and 576/50i for broadcast. I would be surprised if either:

1.) None of this content is ever filtered – this would result in horrendous flicker on all interlaced displays

Wellll ... since it seems to me, that more and more SDTV is showing perceptible, unacceptable flicker these days anyway.
Interactive services (with graphics created in a box sitting on the customer's TV) are the worst offenders, though that doesn't strictly count towards our discussion.

Interestingly, nobody has ever cared about the 1080i flicker, really. This basically comes down to the fact that all 1080i sets are quite recently built, so that it is fully expected for them to show 1080p24 content (this, being the worldwide interchange standard now). And this is basically quite easy to achieve now, since most 1080i-capable TVs are progressive panels, ie LCD or plasma. Any that are CRT, are necessarily going to show 1080i at a quite small size, and therefore have less noticable flicker anyway.

Let's remember that every engineer who's ever tried to build something capable of 1080i, has quite obviously had in the back of his mind, that 1080p is the holy grail, and that a heckuvalot of 1080i content is (very predictably) going to be coming from films. Ever since 1080i was talked about and designed, people have been expecting to cram 1080p into it when possible anyway.

Quote

OR:

2.) All of this content was pre-filtered during the mastering process -  If so this would mean any worldwide networks opting to use the 720p or 576p standards would be broadcasting material that is unnecessarily filtered and therefore missing valuable resolution.

Yeah this is unlikely to be true, as you say. Though even if it was, a bit of resolution reduction on the 1080i master should then, with proper conversion, be no longer noticeable once 720p was created. If you're starting at 1080i50 with a p25 source, and heading towards 720p25, then afterwards there shouldn't be a difference -- regardless of any filtering applied to the 1080i stage.


Bottom line though -- every tape the networks get is either 1080i50 or 1080i60. So the question really comes down to the mastering process. And I say the same applies here as in point 1, where the engineers throw care to the wind when it comes to interlace artefacts, and go all-out for maximum resolution when producing 1080p24 and 1080i60 tapes.

So we're guessing what goes on inside a few buildings in Los Angeles;  I can't speak too much towards the high-end hardware, except to say I've never come across a statement of vertical resolution filtering;  indeed the most I've heard on the conversion process is the symmetrical nature of the conversion, meaning that 1080i60, 1080i50 and 1080p24 are totally interchangable, thus implying complete 1:1 pixel transfers.  As for the pro-sumer end, I know for a fact that there's no automatic conversion in the video editing tools (which can now be used by most content creators on total-HD projects, often at 1080p24 levels).  The only way it's going to occur is by the use of a CRT HDTV for the purposes of proofing and previewing, and presuming they use this instead of a more common LCD HDTV, and also presuming they have cameras that are good enough to make really sharp 1080p24 rather than what most reasonably-expensive equipment produces (which is "rather quite impressively sharp 1080p24" instead), and further presuming they have a good enough eye for the detail and flicker for small-to-medium CRT sizes, then ... THEN, they'd have to be fairly brave to actually introduce a resolution-reducing filter into their production, which actually took away from the detail they've gone to all the trouble to create.  Most of them would pick 1080p24 over 720p24 purely for the detail, and you know what -- they'd already be restricted into delivering via 1080i60 because of the rarity of delivery & broadcast formats that support 1080p24.  They'd be transferring to 1080i completely out of compulsion, and for that they'd absolutely want to prevent tampering with it and hope to heck they & their clients can view it on a system that knows about 3:2 pulldown.

Quote

The only logical conclusion is that filtering is being applied by the TV networks themselves. It may be as simple as option in a Terranex video processor.

Sure, but that assumes they care about the flicker in the first place (but we were talking about that two quoteblocks above here).

Quote

This is further backed up by screenshot comparisons of 24p/25p film based material (video based is another matter) between Seven SD/HD. For 24/25p content, weave de-interlaced 576i and native 576p broadcasts should theoretically look very similar (bit-rate differences aside) if filtering was not applied to the 576i broadcast. However its clear that when looking at screenshots comparisons between some shows (Lost being a prime example) that the progressive broadcast has clearly better resolution. This can actually be quite dramatic. I would say in the order of 10-20%.

Unless there is some other factor affecting the quality of Seven’s SD broadcasts, then I cant imagine where this big difference in quality is coming from.

Ah. I think there is, and I can explain it. Grab a frame of the 576p service, and then "create" a grabbed frame off the 576i service, and you should expect a difference, even when the source was 1080p25 via 1080i50:

The 576p grabbed frame will basically represent a field-drop of 1080i50. Start with 1920x1080, field drop to 1920x540, and scale to 720x576 and spurt out one frame of 576p50 transmission. Repeat with the other field to create the other 576p50 frame, and you've then finished with that film frame (1920x1080).

Whereas ... the 576i grabbed frame actually represents that entire film frame in its entirety. (Remember folks, when you consider how to scale it down to SDTV, you always need to create good 576i50 out of 1080i50 because 50fps motion may be present). The 50fps handling is always in effect -- unless someone at the TV station is actively pressing buttons (or a good processor is involved, which has to handle 1080p50 internally -- that's rare) (or, a tape with direct film-to-SDTV transfer is supplied -- assuming that the TV station can handle dual playout ... well, Nine HDTV can do it for OAR, but this is Seven we're talking about). That means the even rows of the 576i frame only represent the 1920x540 from that film frame, and the odd rows represent the other 1920x540.

So think about it: You're faced with 540 lines to produce each 288 lines on the output. Some of those details are going to get thrown out, most likely. If you have a fine piece of detail that only occurs emphatically on only one of the 1080 lines (which is a =good= thing with 1080p masters ... especially when they come from 4096x4096 Hollywood masters), it will only appear in one set of 540 lines too. When this gets scaled to 288, it might either get thrown out, or averaged across two of those 288. In which case, the detail disappears completely, or is smoothed out across HALF THE SDTV LINES ONLY. This is then weaved in with the other half of the SDTV frame, which, you remember, never contained that detail in the first place. Result? Deliberate haloing, if you watch the 576i as frames on purpose.

Film -> 1080i -> 576i is gunna be messy. But anyone who cares about that will be watching the HDTV service already, and couldn't give a stuff about the 576i service.

The best SDTV result is always going to be the DVD, which apart from having more megabits to play with (in VBR too), will have a 4096x4096 -> 720x576 direct scaling, calculated purely on the frames, ignoring fields completely if they're good about it. Sure, that way you'll get flicker on some TVs, but it'll also look great on computer -- which is something people repeatedly tell you.

It'll also look great on a progressive-output DVD player on your HDTV set.

Quote

There used to be a very good page dedicated to the topic of interlaced filtering at www.progressivescan.co.uk. Unfortunately it seems the site has been removed.

Maybe www.100fps.com is a worthy substitute?

CK.

[ckent]

Ikari, on Jun 17 2006, 04:20 PM, said:

Quote

There's one more idea: If you do the bobbing, like line doubling, but introduce BLACK lines instead of doubling the lines, then first of all you'll end up with something a little like interlacing as seen drawn onto a CRT.

I was thinking the exact same thing: bobbing with interpolated bits replaced with black lines

note: the aniGIF doesn't seem to work very well unsynch'ed on an LCD screen.

That's a great demonstration, Ikari. This is the only way bobbing will ever achieve full resolution. (Just turn up the frame rate all the way).

Remember, the only reason this works, is that there is no rubbish "guessed" pixels being inserted, where the interlacing has originally removed them.

That's why SBS, Seven, ABC, even Ten are rubbish on their upconversion. Only Nine (and Ten's internal-to-HD-studio vision) actually do any processing.

Damn it, even TVs from over 5 years ago can do better "guessing" on 576i material.

CK.

[Darklord]

ckent said:

It seems to me we're basically agreeing on most of this, then  

Seems that way to me too

ckent said:

.. since it seems to me, that more and more SDTV is showing flicker these days anyway.
Interactive services (with graphics created in a box sitting on the customer's TV) are the worst offenders.

True. Although keep in mind STBs/DVD Players can also filter the interlaced output (Stephen and I were discussing this above). However I would agree that some do not.

ckent said:

More to the point, if you compare modern examples to older ones, you'll quickly find that in those days everything was softer and less precise anyway, so without the digital manipulation back then it was impossible to find anything as sharp as you could see nowadays.

Exactly.

ckent said:

Welllll ... interestingly, nobody has ever cared about the 1080i flicker, really. This basically comes down to the fact that all 1080i sets are quite recently built, so that it is fully expected for them to show 1080p24 content (this, being the worldwide interchange standard now). And this is basically quite easy to achieve now, since most 1080i-capable TVs are progressive panels, ie LCD or plasma. Any that are CRT, are necessarily going to show 1080i at a quite small size, and therefore have less noticable flicker anyway.

I’m not so sure. Its true that these days the majority of HD capable sets are digital, but there are still a lot of HD CRTs (particularly RPTV) sets out there that need to be catered for. I’d be surprised if filtering of 1080i content had been abandoned altogether. Particularly in other countries like the US where CRT direct views and RPTVs are still very popular.

I am however only going on old articles and educated guesses. I’m quite happy to admit I have no real evidence to back this up, and to be corrected if I’m wrong.

As far as SD 576i goes I would be almost certain that filtering is still applied to the majority of broadcasts.

ckent said:

Yeah this is unlikely to be true, as you say. Though even if it was, a bit of resolution reduction on the 1080i master should then, with proper conversion, be no longer noticeable once 720p was created. If you're starting at 1080i50 with a p25 source, and heading towards 720p25, then afterwards there shouldn't be a difference -- regardless of any filtering applied to the 1080i stage.

I disagree on this one. Regardless of what resolution you’re converting to, if the original material is softened through filtering, this will translate into a lower quality down-converted picture. Just as in Photoshop when you apply a light softening to a picture, and then down-convert, some of that detail is lost through the softening process.

ckent said:

Ah. I think there is, and I can explain it. Grab a frame of the 576p service, and then "create" a grabbed frame off the 576i service, and you should expect a difference, even when the source was 1080p25 via 1080i50:

The 576p grabbed frame will basically represent a field-drop of 1080i50. Start with 1920x1080, field drop to 1920x540, and scale to 720x576 and spurt out one frame of 576p50 transmission. Repeat with the other field to create the other 576p50 frame, and you've then finished with that film frame (1920x1080).

Whereas ... the 576i grabbed frame actually represents that entire film frame in its entirety. (Remember folks, when you consider how to scale it down to SDTV, you always need to create good 576i50 out of 1080i50 when 50fps motion is in effect). The 50fps handling is always in effect -- unless someone at the TV station is actively pressing buttons (or a good processor is involved, which has to handle 1080p50 internally) (or, a tape with direct film-to-SDTV transfer is supplied -- assuming that the TV station can handle dual playout ... well, Nine HDTV can do it for OAR, but this is Seven we're talking about). That means the even rows of the 576i frame only represent the 1920x540 from that film frame, and the odd rows represent the other 1920x540.

So think about it: You're faced with 540 lines to produce each 288 lines on the output. Some of those details are going to get thrown out, most likely. If you have a fine piece of detail that only occurs emphatically on only one of the 1080 lines (which is a =good= thing with 1080p masters ... especially when they come from 4096x4096 Hollywood masters), it will only appear in one set of 540 lines too. When this gets scaled to 288, it might either get thrown out, or averaged across two of those 288. In which case, the detail disappears completely, or is smoothed out across HALF THE SDTV LINES ONLY. This is then weaved in with the other half of the SDTV frame, which, you remember, never contained that detail in the first place. Result? Deliberate haloing, if you watch the 576i as frames on purpose.

Film -> 1080i -> 576i is gunna be messy. But anyone who cares about that will be watching the HDTV service already, and couldn't give a stuff about the 576i service.

The best SDTV result is always going to be the DVD, which apart from having more megabits to play with (in VBR too), will have a 4096x4096 -> 720x576 direct scaling, calculated purely on the frames, ignoring fields completely if they're good about it. Sure, that way you'll get flicker on some TVs, but it'll also look great on computer -- which is something people repeatedly tell you.

Ok. My head REALLY hurts after reading that. Had to read it twice just to take it all in, but I think I know where you’re coming from. I guess another basic way of putting it is to say that because Seven HD basically treats all fields as frames and bobs 540i fields directly to 576p, there can often be more detail in the following frame due to differing line positions from one frame to the next (unlike 576i-> deinterlaced to 576p which will have full resolution 576p frames, but each frame is followed by an identical repeat). There you go Ikari! A process where we can both agree bob actually does result in more detail! . The negative is that you’ll get the same bobbing up and down of detail (plus ugly aliasing), and even in the case of film it will lack the solidarity of true progressive 576/25p.

ckent said:

Maybe www.100fps.com is a worthy substitute?

There’s some decent info there, but not a lot in the way of filtering of interlaced broadcasts. progressivescan.co.uk had a page dedicated to this exact topic. Wish I could track it down.

[pneu]

Darklord, on Jun 23 2006, 04:49 PM, said:

I guess another basic way of putting it is to say that because Seven HD basically treats all fields as frames and bobs 540i fields directly to 576p, there can often be more detail in the following frame due to differing line positions from one frame to the next (unlike 576i-> deinterlaced to 576p which will have full resolution 576p frames, but each frame is followed by an identical repeat). There you go Ikari! A process where we can both agree bob actually does result in more detail! .

Are we talking about 1080i bobbed to 576p, compared with 1080i downscaled to 576i and then weaved?

If we are, then I'd have to say the 576p bobbed version should really be the same resolution as the 576i version weaved.  Because with 576p, the maximum vertical resolution that can be obtained is limited to 576 lines, and 576i weaved achieves the full 576p resolution already.

Having said that, the bobbed 576p version still carries all of the 540 line fields from the 1080i version (ignoring horizontal resolution), but in order to achieve more than 576 lines worth of vertical resolution you'd have to manipulate the frames in such a way that would interleave the lines from the first frame with the next one... this is what I did with the Sunrise example - took 2 succesive 576p frames (courtesy of DavoNogo ), interleaved them together and achieved roughly twice the vertical resolution (well, in theory anyway...)

[Darklord]

Ikari, I think what CK is saying (its not 100% clear and I believe further clarification is in order) is that because the Seven SD feed is often down-converted from the HD feed, the SD 288 line fields are converted from 540p frames (after down-conversion from 1080i). Hence uneven line distribution and further interpolation taking place? So even in the case of film sources those fields are going to lose further resolution in the conversion process. I guess you could argue this IS a form of filtering!

The big difference in resolution between Seven SD and HD obviously comes with material that is shot in native 1080/50i and converted to 576p for broadcast as you then have 50 unique motion updates a second comprising of 540 unique lines or resolution (vs 288 unique lines of resolution and 50 unique motion updates of SD 576/50i). However it seems CK is saying there can still be a difference with film based 24p/25p material due to Seven’s conversion process.

I’d still argue that some from of filtering is taking place (direct or indirect), as if done properly as you said 576/50i should be able to carry the same data as 576/25p in the case of film based material (video comprised of 50 unique motion updates is an entirely different story – this is where native progressive broadcasting shines).

Arghh.. I don’t know.. All I do know is my head hurts

Anyway I'll have to leave you guys to continue this interesting discussion, as my ADSL gets disconnected tomorrow (I am moving house next Saturday) then I am leaving to go overseas for 5 weeks! As a result I won’t be responding to many threads before I leave (and none at all between July and Mid August).

I look forward to seeing how much this thread grows while I’m away (or perhaps I should say I’m dreading to see how much it grows... )

[pneu]

Darklord, on Jun 23 2006, 11:31 PM, said:

Arghh.. I don’t know.. All I do know is my head hurts

Mine too!  

ckent, could you clarify what you are suggesting 7 is doing when they convert 1080i/50i and 1080i/25p to 576p?  Last time I checked (it was a while ago) 1080i/25p programmes (eg. The Great Outdoors, Blue Heelers etc.) on 7HD were not showing signs of bob de-interlacing (no bob flicker) and I assumed they were using weave fields to 1080p before downscaling to 576p.  At least, that's what they should be doing.

And for standard def, downscaling 1080i50 to 576i50 should require no more than simply scaling each 540 line field to 288 lines.  Alternatively, for 1080i/25p sources, if they weave fields to 1080p before downscaling to 576p, then split that into 2x288 line fields, it might yield some more resolution...I'm not sure (was this what you were getting at?).

[pneu]

ckent, I think I understand what you're saying now - that down scaling each 540 line field to 288 lines, and then weaving 2 of those downscaled fields to 576p, results in a lot more averaging of pixels overall, whereas a single 540 line field upscaled directly to 576p undergoes much less scaling, so it is overall more "faithful" to representing an original 540 line field.

You could very well be right about this.

I think it's time for some more photoshop comparisons!  

[pneu]

Fields downscaled, weaved to 576p
Single field upscaled to 576p
Entire 1080p frame downscaled to 576p

(original test pattern here)

You were spot on, ckent.  Definite blurring when fields are scaled seperately.

[ckent]

Yep. You'll see the exact same symptoms on:

- Nine's Today show on the SD service: All of the on-screen text does this, eg the weather on the bottom strap
- Seven evening news on the SD service: Text inserted on SD tape, before the HD studio
- The Sunday show replay, as seen on Sky News: Check all the logos, etc, very easy to see!
- Prime Wollongong, which I could receive a few months ago, appears to be taking an HD feed and using it for both SD and HD services, creating a substandard SD service for all shows.

Of course, none of this would ever be a problem if all conversions were done via 'p' progressive first. 1080i to 576i has to go via 1080p to be done properly (not 540p). Same goes for 576i to 1080i, should be via 576p (not 288p), but most of us probably already understood that from seeing the performance of upconverted content on HD channels, where Nine HD is noticeably crisper.

It's always easy when you have film, producing masters for 576i or 1080i, because you can just treat everything as progressive all the time. It's easy to scale a 4Kx4K master or 1080p24 master and just dump it to 576p25, encoded as MPEG-2 576i. Looks great, too.

The tricky bit of course is handling 1080i50 with actual 50fps motion. Well, the folks at the World Cup in Germany can do it. The quality of the SD 576i output is absolutely lovely. Now imagine the 1080i quality.

CK.

[pneu]

ckent, on Jun 27 2006, 04:29 AM, said:

The tricky bit of course is handling 1080i50 with actual 50fps motion. Well, the folks at the World Cup in Germany can do it.

How would they do this?  Per-pixel de-interlacing to 1080p, then downscale to 576p?  Or just using better scaling algorithms?