HV20 - a real progressive sensor or not?
 

My friend shooted in 25p mode with the HV20 in Seychelles.
The video isnt smooth, very eye strain for me. I think its because lack of ND filter, and the camera use very short shutter speed.

But my real problem is, that not every frame looks progressive! Yes! Not every. I import the native m2t file into Edius4 1440x1080 25p project, and set the file properties to progressive (because edius think its a 50i file). When i go from frame to frame i found this 2 interesting frame!
(indicated with green arrow)
http://www.relaxvideo.hu/hv20a.jpg
http://www.relaxvideo.hu/hv20b.jpg

this looks like a deinterlaced frame from an interlaced video, doesn't?
but the next few frame is ok (one tree, one rock, without shadow)
So periodically hv20 insert a deinterlaced frame or the whole cmos sensor isnt a real progressive one?

This is only for PAL model, i dont know the NTSC version, but like to hear about it.

Hi Prech. From what I see, that could be motion blur rather than deinterlaced video.

Richard

I think I agree w/Richard. That doesn't look like interlace artifacting.


-A

yes, it does, every other frame looks normal.
One blurred tree trunk, not 2 shadowed..

I work in video area for many years now, and from my experience, this looks like a deinterlaced frame.

If its just motion blur, then why do i see two trunk from one tree?
I upload the previous and the next few frames, you will see the difference:
http://www.relaxvideo.hu/hv20

I have seen this ghosting as well on faster moving shots at 24p (and you are in more authorative position as you use 25p, without a need for pulldown removal). It looks like resampling to me.

Resampling may result blurry image, but why ghostly?
I know it has to be resampled from 1920x1080 to 1440x1080, but when you squeeze horizontaly the full progressive frame, why appear this edge ghost effect?
It seems to me, that hv20 can readout the sensor only in interlaced mode, but in 25p mode doesn't have to wait 1/50 sec for the next readout, like in normal 50i mode. Instead if it finish the upper rows, immediatly goes for the lower rows, and its take time because the rolling shutter effect.
What else can explain this?

Hi Prech. I have an HV30 but to be honest I have not used it very much so far and have not played around with all the modes yet.

But from what I know, when you set the camera to Cine mode, it is possible that the camera will adjust the shutter speed automatically while you are shooting. So I am wondering if you are seeing motion blur in some frames but not in others because the camera has selected a slower shutter for those frames? Is it possible?

Richard

No, it is because it was a handheld shot. sometimes the pictures is steady (no motion blur), another time on fast moving, motion blur exist. This is normal.
What would be the reason to change the shutter speed for an existing landscape and light situation?
And even if camera change the shutter speed, that doesnt explain the two trunk. dont you think?

Look at the TV in the image...
 

This is a single frame from a clean 24P-deinterlace process, HDV. Obvious motion blur and a slight lean, and I'm going assume the shutter was 1/24 due to the low-light.

The bright areas of the image have a blur as the camera was in motion, but I'd think the TV's 'multiple exposures' are merely due to the fact it's scanning at 60i. The LCD monitors - also bright - blur as one would expect - smoothly.

It's a CMOS sensor, and I expect it to be true progressive. The HV20 is a year old now, and this topic has been throughly discussed here.

Hi Prech. I understand what you are saying about the shutter speed, but it is also possible that the lighting conditions are just on the threshhold of what would make the camera select a different shutter speed. I'm not saying this is likely, just mentioning the possibility.

The 2 trunks thing does look strange, but it doesn't look like a deinterlacing issue to me. Do you happen to know what the shutter speed was for this shot? Also, have you seen such an effect when the shutter speed was fixed?

Richard

Richard, unfortunatelly i wasnt in Seychelles yet.
My friend was there.
If the light condition change a little bit, i think the camera first try to adjust the iris, not the shutter speed. (on prof. TV camera, shutter speed is always 1/50).
but anyway, again my question: if there is a shutter speed change, why do is see on this frame two trunk?
What else when not a two different field -> one full frame conversion?

regards

Hi Prech. This is a clip taken with an HV30 in 25P progressive mode.

http://www.jaegersing.com/25P_Test.m2t

I opened it in Vegas and every frame looks progressive to me. The shutter speed is around 1/250s so there is little motion blur to confuse matters.

Maybe you can take a look at it using Edius and see if you get any strange frames?

Richard

Sorry I had to take down this clip because it exceeded my bandwidth allocation. Oops!

yes!
it looks a real progressive material
but with hv30

Yeah, I wasn't sure how useful it was to compare this way round. Maybe you could post a short section of your "double-image" m2t and let me take a look in Vegas?

Richard

The 24fps progressive images of your camera are stored within a 60fps interlaced envelope. To extract the original frames, you need to pass your footage through a process known as 3:2 pulldown.

Here's one way to do it:
http://eugenia.gnomefiles.org/2007/0...-24p-pulldown/

It was wrong of you to resize the image. By doing so, you ended up blending the interlaced frames. Also, the video will be really stuttery on playback. More stuttery than normal 24p.

It's either you leave the image at 60i and output it to a DVD which results in a perfect image on a real TV set or you go through the pulldown process discussed in the link supplied above, or some other method.

Seun,
your answer is for me?

If yes, you're wrong. I know the 3:2 pulldonw process.
My frined, don't shoot in 24p, because PAL model only has 25p, and with 50i format, it doesnt need to remove pulldown. Only with ntsc model.

So, in PAL models every frame (2 field) from 50i video, that was shooted in 25p, has to be a real progressive frame without any magic.

Look at Richard's sample, it is from Hv30, and 25p, not 24.

Do you have any other examples? This one seems pretty odd. If it only happened with this shoot, then who knows why the tree look strange?

It could be lighting, it could be some setting that was overlooked, it could be your particular camera, it could be the motion of going handheld, it could be image stabilization was turned on, it could be many things particular to this shoot. Much less likely is that you've discovered that the HV-20 has an interlaced CMOS sensor.

Look at more examples. And post them if you see the same problem.

I believe Prech has a good understanding of video and knows what he is talking about.

The HV20/30 video is progressive but it's also PsF, so potentially there are interlace factors to take into account. What I mean is that the video captured through firewire is not simply 25 progressive frames per second as in the case of the Canon XH-A1, but rather it is 25 progressive frames segmented and distributed across 50 fields. If the NLE interprets this footage wrongly (may or may not be due to wrong project settings) then I can see the possibility of some deinterlace issues.

Even though I said previously that Prech's example does not look like a deinterlace issue to me, I am prepared to consider the horrific possibility that I am wrong. :) The most important thing, for me at least, is to know what's actually going on here.

Richard

I'm in NTSC land, so I deal with the 3:2 pulldown. But you're saying that the PAL version uses essentially a 2:2 pulldown to embed 25P in 50i. Makes sense, I just hadn't really considered that PAL uses a pulldown as well.

I think you may be on to something with the compression and subsequent decompression blending fields from different frames. This is a problem with DV casusing every third frame to have distortions; which is why the DVX comes with the Advanced Pulldown option. But this should not be a problem with HD.

But I sounded somewhat critical in my post, I just don't see how it can be the sensor that is not progressive. Of course I could be wrong, but it just goes against everything I've seen, read and heard. So that's why I suggested running more tests and double checking and listing the settings.


P.S. Actually, come to think of it, this problem could very well be caused by how the frame grabs are performed. A lot of players will try blending fields even with progressive images when frame grabs are done. For example, I've found VLC introducing ghosting on frame grabs when none exists in the raw footage.

So along the line of your NLE ?, how were there these frame grabs taken?

Peter,
25p PAL HDV doesnt need pulldown, just to interpret the footage to progressive.

"how were there these frame grabs taken?"

read #1 post

I wasn't sure you used Endius for the frame grab as well as viewing, but after re-reading your post, I see that's what you did.

In all seriousness, can you please explain to me how PAL doesn't use a pulldown? In NTSC, progressive frames are repeated 2, then 3, then 2 and so on into the 60i fields.

But with PAL, aren't frames repeated twice to fit into the 50i fields. How is this not a pulldown? It's not as annoying as the 3:2 pulldown, but isn't it a pulldown nonetheless? Progressive frames are being repeated to fit into a different timeline (used for televison broadcast), isn't that what a pulldown is? Or are you saying the frames have to be repeated at an uneven cadence to be officially called a pulldown?

Back to your problem. Have you seen it with any other footage? Have you tried loading the footage into a different NLE?

i ask my friend, to give me another fast moving videos from hv20.
(because i personally have hv10)

60 is not a multiplier of 24, thats because ntsc model need pulldown.
but 50 = 25x2, so we are lucky. not need to magic, just say to NLE, this 50i footage is in reality a 25P footage. That is enough.

Prech,

I realize that 60 and 24 need a common multiple, hence the 3:2 pulldown. But just to clarify things a bit, the PAL version does indeed use a pulldown. It's just an easy one to deal with, 2:2. But the process of putting the progressive frames into interlaced fields for broadcast IS the pulldown. And both PAL and NTSC versions do this.

Removing the pulldown has to be done in NTSC land to edit properly. In PAL land apparently you can leave the pulldown in there w/o any problems. But both have a pulldown.

As for the problem you presented, more footage or info on the settings your friend used would be helpful.

All,
I'd put my money on the optical image stabilizer having an occasional jitter. I'm personally not very impressed by its performance.

Less likely:
- motion blur
- the NLE trying to deinterlace of trying to combine odd&even fields from subsequent frames instead of combining fields from the same frame.

A nice test to check whether your NLE keeps the fields clean is to run some 50i interlaced video through it. It should look like this:
http://www.artech365.com/winvdr/images/invers1.jpg

See: odd&even fields perfectly untouched. From the HV20 25p progressive mode, the fields should match because they were shot at the same moment.

regs,
Pieter