How PsF video from the V1 is different than "p" or "F" video

[Piotr Wozniacki]

I have noticed something interesting with my 25PsF video that I don't understand. When I play it back with VLC with deinterlace off and take a snapshot, the resulting picture is lacking a lot of lines at the bottom (it can't even be displayed by some software, like Nero Photosnap Viewer - but you can display it with the Windows own Picture and Fax viewer, or Paint). When I switch deinterlacing on, the snapshot is complete (it's also complete from the V1E interlaced video, or from the Canon A1 progressive and interlaced).

Can somebody explain why that happens? Perhaps the missing bottom lines have something to do with the true 25PsF resolution?

Here is one such "incomplete" snapshot (to upload it, I had to convert from png to jpg)

UPDATE: the previous uploaded pic didn't have full resolution, this one has - and is reported as 1920x1088 here.

UPDATE:Unfortunately, the second one also gets scaled down when uploaded - hence jagged edges which are perfect in the original; don't know why...I also added a snapshot from the same clip with bobbing=on; the scan is complete but aliased lines are clearly visible.

UPDATE:Finally, both uploaded snapshot *are* full resolution after all; when downloaded with IE they must be maxed using the maginifying glass cursor tool. Then you can clearly see the difference: the first (left) picture is not de-interlaced and has no jagged edges, while the second (right) one is bobbed and has plenty of aliasing. The questions remain:

- why on earth cannot we play back (through HDMI or component) with the quality as in the first picture, ie. without deinterlacing?

- what does the missing bottom lines phenomenon come from?

[Steve Mullen]

Originally Posted by Bob Grant

Piotr, the V res is not too high. With 25PsF or 25p 1080 you really can have 1080 lines of V res.

You can record up to 1080-lines, but the question is what comes from the EIP. Given the number of CCD pixels in a column (540) there is an upper limit to the number of lines of V rez. that should be captured. (Way less than 1080.) This number is determine by the Nyquist value. A low-pass filter SHOULD remove all information above this number before the A/D to prevent aliasing.

When Piotr says too much, he means raising the filter higher than it should be. This allows more V detail which is why we see twitter. Aliasing is extra NOT REAL detail, and is visible as the dancing noise.

[John Bosco Jr.]

Originally Posted by Steve Mullen

You want data -- the over 5000 hits recorded by the site is far more "data" than your opinion or spots'.

You also ignore the fact the humans can make judgements that work in certain cases better than instruments. In fact rez "measurements" are made by humans. Do you consider them not "data?"

Because Piotr is trying to choose between the two camcorders. Read the past posts. Even better, read the tread topic. What camcorder has 24F?

And, if don't find anything of value in this thread -- why are you reading it? Why post something that has NOTHING to do with the topic. Do you think folks are interested in a post that has only your opinion of the thread?

Don't like a thread -- don't read it.

Well said... Of course, we should not get into a battle of V1 vs A1. Obviously, some users have had bad experiences with the V1 and are very bitter. However, if you try to understand the camera instead of bash it; you might begin to like it. Piotr is a prime example of a user who is trying to find a solution to a so-called problem with the V1E. Misinformation about any camcorder doesn't do anyone any good.

In my opinion, and this is only my opinion. The picture of the V1 and HVX 200 looks better than canon's A1/G1. So my decision for a camera is between those two models. However, I'll wait until NAB to make a decision. I'm hoping Sony will come out with an update to the V1 with 1/3rd inch CMOS sensors. After all, CMOS is the better sensor for High Definition. Why you ask?

Have you ever wonder why it seems that HD video is more noisy than SD video from CCD cameras? Well, It is because the S/N ratio of CCD-based cameras degrades ~3 dB per octave. For example, if a CCD camera has a standard definition S/N ratio of 62db, in high definition it will be 54db. Of course, you can apply more noise reduction but at the expense of lower resolution (1).

CMOS cameras, on the other hand, are capable of reducing noise and maintaining resolution because the noise-setting bandwidth is at the pixel amplifier rather than the video output amplifier, and since the pixel-based amplifier’s bandwidth better matches the imager sampling frequency, the CMOS output buffer’s noise is usually negligible, and the result is a higher S/N ratio (2).

These facts were obtained from an article on broadcasting high definition television called, "CMOS vs. CCD: Changing Technology to Suit HDTV Broadcast." Hopefully, you find it useful.



Additional Sources:

1. K. Mitani, M. Sugawara and F. Okano, “Experimental Ultrahigh-Definition Color Camera System with Three 8M pixel CCDs,” SMPTE Journal, April 2002.

2. M. Loose, L.J. Kozlowski, A.M. Joshi, A. Kononenko, S. Xue, J. Lin, J. Luo, I. Ovsiannikov, J. Clarke and T. Paredes, “2/3-inch CMOS Imaging Sensor for High Definition Television,” 2001 IEEE Workshop on CMOS and CCD Imaging Sensors, June 2001.

[Bob Grant]

Originally Posted by Steve Mullen

You can record up to 1080-lines, but the question is what comes from the EIP. Given the number of CCD pixels in a column (540) there is an upper limit to the number of lines of V rez. that should be captured. (Way less than 1080.) This number is determine by the Nyquist value. A low-pass filter SHOULD remove all information above this number before the A/D to prevent aliasing.

When Piotr says too much, he means raising the filter higher than it should be. This allows more V detail which is why we see twitter. Aliasing is extra NOT REAL detail, and is visible as the dancing noise.

Sounds like a great theory and one that I had hinted at some time ago except it doesn't apply!
The A>D converter is converting lines, not columns so it's only a factor in the H direction.
All else being equal, same sensors etc when they're being clocked interlace V res is reduced by 30% as line averaging is applied to contain line twitter. This improves S/N by 6dB. Note this is line averaging which is quite different to a low pass filter.
When clocked progressive line twitter is impossible (assuming it's displayed corrrectly!), so line averaging is turned off. You get the full res but 6dB less S/N.

The above is from a Panasonic white paper, sorry I can't find a link to it anymore.

Bottom line is a progressive scan video camera, no let's stop calling them that, a digital film camera, takes a sequence of still images, film cameras, RED, SI-2K, D20 etc none of them will employ line averaging or limit V res in any way. You want to display moving images at full V res from them on interlaced displays, well without processing you're going to have the potential for line twitter.
If you doubt what I'm saying take my 1080 res test image, my P displays have no problem with this, a full 1080 lines of V res. Try displaying a HD res chart, on a display that's displaying P correctly you'll get to 1000 line no sweat, on an interlaced display at around 700 line it'll fall apart.
The question might be, should Sony have done something to limit V res so the progressive scan images would display correctly on legacy interlaced displays, fair question. Are the other HDV cams that record P doing this in camera, I suspect so, I've not read of one offering 700+ lines of V res.

I'm a few days away from Sony's Full HD presentation, be interesting to see what they have up their sleeve, maybe they have a HDTV that'll display the images from the V1 correctly.

[Piotr Wozniacki]

Bob, just to make it clear - my monitor can display your chart with no problems; thanks for this great test!

[Bob Grant]

Originally Posted by Piotr Wozniacki

Bob, just to make it clear - my monitor can display your chart with no problems; thanks for this great test!

Great, step 1!

So on the same monitor, same everything do you have issues with twitter on images from the V1E?

If anyone's interested the ISO 12233 res test chart should come in handy for seeing what you monitor(s) are doing. Digitise it to 1080p HDV correctly framed, no AA filtering. I'd post a copy but it might be copyright, A Google will find a vector PDF version of it.

The first step in testing anything should be testing your test equipment, yes this is measurabator stuff. Sorry but I spent a lot of time working around a standards lab, this was drummed into me relentlessly.

So your first port of call should be to take the camera out of the test. When and only when you know your test system is upto the task at hand then bring the camera's images into the test system.

[Piotr Wozniacki]

Originally Posted by Bob Grant

Great, step 1!

So on the same monitor, same everything do you have issues with twitter on images from the V1E?

YES - everything the same, the monitor fully capable of clearly showing your chart with all lines distinguishable - yet the 25PsF is twittering when fed from the camera via component, or played back from a captured raw file with VLC bobbing.

No twitter at all when fed from camera compressed through firewire, or played back from VLC with deiterlacing off.

I'll repeat what I have already stated many times: the component input on my monitor (as well as the component and HDMI inputs on the Bravia HDTV I tested) are doing to the 25PsF from camera exactly what bob deinterlacing does to a raw captured file played back from the HDD.

[Bob Grant]

Originally Posted by Piotr Wozniacki

YES - everything the same, the monitor fully capable of clearly showing your chart with all lines distinguishable - yet the 25PsF is twittering when fed from the camera via component, or played back from a captured raw file with VLC bobbing.

No twitter at all when fed from camera compressed through firewire, or played back from VLC with deiterlacing off.

I'll repeat what I have already stated many times: the component input on my monitor (as well as the component and HDMI inputs on the Bravia HDTV I tested) are doing to the 25PsF from camera exactly what bob deinterlacing does to a raw captured file played back from the HDD.

OK
but just to be really clear. If you're feeding that monitor by DVI from a PC that's very different to feeding it via HDMI or component.
What happens if you print that test image to tape and feed the camera to the monitor via HDMI or component playing it back?

See when a monitor is connected via DVI from a PC it's refresh rate, de-interlacing etc is being controlled by the PC. My Dell 2407 is a good example, it displays the 25PsF perfectly via DVI from a PC out of Vegas, feed it component from the camera and yuck.

What you've seen using VLC ties in exactly with this, VLC lets you emulate what HDTVs are doing, mostly wrongly too.

[Steve Mullen]

Originally Posted by Bob Grant

Sounds like a great theory and one that I had hinted at some time ago except it doesn't apply!
The A>D converter is converting lines, not columns so it's only a factor in the H direction.
All else being equal, same sensors etc when they're being clocked interlace V res is reduced by 30% as line averaging is applied to contain line twitter. This improves S/N by 6dB. Note this is line averaging which is quite different to a low pass filter.
When clocked progressive line twitter is impossible (assuming it's displayed corrrectly!), so line averaging is turned off. You get the full res but 6dB less S/N.

I'm not a full beliver in the "theory" for exactly the reason you said -- the A/D is sampling along rows.

But, we do see aliasing on rows, even with the V1U. So something equivelent must be happing as cameras are "sampling" an image vertically. Nyquist still must apply. I just don't know what or how.

Moreover, the E really does have significantly more aliasing vertically than does the U. It's not simply that the camera does not filter enough. My U works exactly like you describe. But, Piotr's does not. He is recording crap in the same 1080 HDV that I have. Something is fundamentally different -- so inherently different that Sony could not simply "adjust" it. Therefore, I think the "theory" is correct--except it's not the A/D.

The last question I'm waiting for an answer to from Sony is: does the V1 do row-pair filtering for interlace. In the past it has always been done. But, the V1's sensitivity does NOT change by 6dB. This is a clue it may not be used. If it's not, a low-pass filter can be used in or before the EIP. But, no mtter HOW it's done-it works fine on the U models. So it is a differentiating factor.

[Bob Grant]

Aliasing happens when a frequency above 0.5 the sampling frequency is allowed to get into the converter. For example with a A>D converter clocked at 1Khz a 1,001Hz signal will produce an ouput from the converter of 1Hz.

For this to happen in the V res, even though the converter isn't working on rows there might be an issue. This camera is employing 3 chips and producing a Y signal by interpolating them, quite possibly that signal contains more than 1080 lines of V res and certainly that could cause aliasing problems.

I think though we need to be careful, line twitter is only a problem with interlaced displays, aliasing can be seen on both I and P displays.

I can see this issue with my 1080 line image. Using Vegas and other programs to downconvert to 720 I can get either a set of fat lines, what looks like graduated fills of varying frequencies etc or a screen of grey. The screen of grey is what I should get but I only get that using bicubic interpolation (Best), all the aliasing problems occur using nearest neighbour (Good).

So yes, potentially you might be onto something with an idea you proposed a few days ago. The camera might be internally creating an image with more than 1080 lines and sampling that into 1080 lines. If not done carefully aliasing could occur. This issue will only occur with an image containing enough detail of course. I believe it's been known to occur in the SD world by using HD glass on a SD camera.

[Steve Mullen]

Originally Posted by Bob Grant

This issue will only occur with an image containing enough detail of course. I believe it's been known to occur in the SD world by using HD glass on a SD camera.

A new firmware load enable the FireWire output from my HD cable box. I can record to D-VHS both 1080i and 720p programming. (No copy protect is on all 16 HD channels!)

And now I can switch my HDTV between the Cable box and D-VHS deck. Both feed analog component. The cable box is obviously softer. The detail from the DVHS VTR is amazingly high -- too high -- and looks like Piotr's S=8 but without the twitter.

So Piotr's questions about METHOD may be unnecessary. It's clear the same TYPE can be very different depending on the analog output (and input) circuits. Even when HDMI is used, unless the HDTV is ALL digital, there is a D/A and low-pass filter on its output.

[Piotr Wozniacki]

Originally Posted by Bob Grant

OK
but just to be really clear. If you're feeding that monitor by DVI from a PC that's very different to feeding it via HDMI or component.
What happens if you print that test image to tape and feed the camera to the monitor via HDMI or component playing it back?

See when a monitor is connected via DVI from a PC it's refresh rate, de-interlacing etc is being controlled by the PC. My Dell 2407 is a good example, it displays the 25PsF perfectly via DVI from a PC out of Vegas, feed it component from the camera and yuck.

What you've seen using VLC ties in exactly with this, VLC lets you emulate what HDTVs are doing, mostly wrongly too.

For now, I have created (in Vegas) a 1080/25p MPEG-2 clip out of your chart and played it back with VLC. As long as interlacing is off, it plays perfectly (of course the lines are not as clearly separate as with a still picture, yet distinguishable). But with bobbing on, the flickering is so awful my eyes can hardly watch it for more than a couple of seconds!

Bob, what's the point of printing it back to tape - I can only do it as 1080/50i anyway? Or am I missing something?

UPDATE: I have rendered another clip in Vegas, this time using the minimum possible Gaussian Blur in vertical direction; while the flickering is easier to watch, the lines are no longer distinguishable. No big surprize!

[Piotr Wozniacki]

I would once again like to express my gratitude to Steve and Bob (first of all - there were others whose input I should also have mentioned), who - I think I can say it safely now - have helped us all establish the V1E/P cameras are not guilty of the "flaws" with their 25PfS video presentation. If I can recreate this flaw with a 1080-lines chart, it's clearly a display device problem (or more specifically: the way a display device is treating the 25PfS signal). The only thing that beats me is how Sony could have released a camera that uses a format which, for the sake of all interlaced playback equipment backward compatibility, is actually not compatible with most up-to-date HD viewing devices. And even worse, instead of informing the community about this fact, they advise to simply turn the sharpness down to 3. Dear Sony, this is not a professional attitude! Turning down sharpness is against the very idea of HD video, and it doesn't solve the problem - a slight line twitter is still visible.

[Bob Grant]

Originally Posted by Piotr Wozniacki

For now, I have created a 1080/25p MPEG-2 clip out of it and played it back with VLC. As long as interlacing is off, it plays perfectly (of course the lines are not as clearly separate as with a still picture). But with bobbing on, the flickering is so awful my eyes can hardly watch it for more that a couple of seconds!

Bob, what's the point of printing it back to tape - I can only do it as 1080/50i anyway? Or am I missing something?

Sorry, should have included a warning about staring at it for too long :)

Open a 1080 50i HDV project, drop image onto T/L and stretch it out a long as you need it for. Render to m2t file and PTT.

The Sony Vegas guys have told me that what will come out the HDMI and component ports of the camera will be identical to what you get for PsF. The V1 is only capable of outputing fields on both those ports.

That test image was created in Vegas BTW.

The point of it, perhaps none.
When you say you played the clip back in VLC, how is the monitor connected?

DVI, Component or HDMI?

What I'm getting at is the monitor may well react differently when it's fed via those 3 different paths. DVI is a pure digital connection, the screen refresh is set by the PC and pixels are mapped one to one, Component is analogue video and HDMI is digital video. By my understanding with the last two the monitor has to process the signal to display it.

I suspect what you're seeing with VLC is pretty much what you'll see going to the monitor via HDMI or component from tape, the monitor is bobbing the image creating the flicker.

[Piotr Wozniacki]

Bob, my ATI graphics card / Fujitsu-Siemens monitor combo enable me to playback (with VLC, from an NLE, or any software that uses video overlay in general) with *BOTH* the digital DVI and/or the analogue Component. There is no much difference between them, apart from the fact that the component HDTV output from the graphics card is more dynamic ("punchy") than the DVI (but of course they can be tuned to match in the ATI Catalyst control center).

So, when I say it's OK with deinterlaing off and awful with bobbing on, I mean both interfaces - the DVI and component.

UPDATE: After a second thought, I must add that the above is a bit more complicated, actually. The thing I'm sort of uneasy about is: why do I need to enforce bobbing in VLC to recreate the twitter even through component, which itself should be doing to the captured 25PsF clip exactly the same it's doing to the live V1E output when fed through it? It beats me...