[Toke]

[Jan Crittenden Livingston] “you say that the reason you didn’t understand it is because I didn’t explain it very well.”

Now you are cutting the corners again. I said I didn’t understand OR you explained it wrong or just eplained other things. Usual thing in conversation, when this happens, you would check your own data and try to tell it in a way that I would understand. You chose not to.

[Jan Crittenden Livingston] “…that somehow the spatial offset is a bad thing.”

I think that spatial offset is a great thing as long as we are recording in YCbCr-more.
Without spatial offset high resolution and high sensivity cameras with this small ccd size and price range wouldn’t be possibe. And here Panny has done just right, unlike many others.

And if I would think that spatial offset is a bad thing, why couldn’t we have a friendly conversation with that? I’m here just for the knowledge and I hope that we could also question Panny decisions without you getting angry.
Btw, there’s lots of people that think that spatial offset is a bad thing, because it uses averaging and therefore it gives you a bit more details, but all details are a bit more blurred. This debate is quite like 1-ccd vs. 3-ccd, because with spatial offset 3-ccd system acts quite like 1-ccd system.

[Jan Crittenden Livingston] “I have said that it was not scaling as the information that the red and blue chips see is information that the green one can not see. It is an additive process, you come back and say it is scaling.”

RBG-chips always see what the other two cannot see, because each of them can only see one color.
With spatial offset (in this case) one color sees both spatially and colorwise different info.
Problem is that in the final picture all color pixels has to be spatially aligned, so they have to re-calculated. This can be done in digital domain (raw-conversion) or like in this case in analog domain.

Spatial offset is an additive process in a sense that you get better luminance resolution to the end picture, but pixels are also a bit blurred because all the averaging. And the downside is that color components’ pixels are also a bit more blurred, because they have to be aligned back to same registration, but they don’t get any additional resolution.
In mathematical sense, as a whole, it isn’t an additive process, because you still get the same amount of data out of ccd’s.
So basically with spatial offset, you get more resolution to luminance and loose some with chrominance. And that suits well with component video recording (and also human vision).

[Jan Crittenden Livingston] “Toke:’Constructing the final YCbCr-image from offsetted rgb-chips is anyway done in after ad-converter, isn’t it?’
No it is not.”

You are really an ace with short answers!
So the whole re-aligning is an analog process and a final YCbCr-image is then digitized.
So for the color and gamma corrections, the image is then re-converted to RGB in DSP?

Generally re-aligning could be done with better quality digitally, because then you could use all neighbor pixels with averaging, as in analog stage you can average only with neighbors in the same row (line).

Btw, Jan you said earlier that most of 2/3″ cameras and especially SD cameras use spatial offset.
Can you give couple of examples what cameras?

2/3″ cameras usually have more pixels in their ccd’s than they record, so I can’t understand what do they benefit from spatial offset. The whole idea is to get more luminance resolution and those cameras already have more than they can record.

[Toke]

[Jan Crittenden Livingston] “Actually you are missing something. An understanding of 1. the fact that the CCD is an analog device. 2. how spatial offset works.”

Well, this wasn’t so friendly. I’ve said many times that I know ccd’s are analog and it wouldn’t make any difference regarding the spatial offset if we were talking about “digital” cmos chips.

Constructing the final YCbCr-image from offsetted rgb-chips is anyway done in after ad-converter, isn’t it?

It would have been nicer if you would have pointed out those things from my explanation of spatial offset that you think are wrong. You are not giving any arguments or explanation to support your statement.

But I’ll wait for Tosh’s blog then…

[Toke]

[Michael Brennan] “So it isn’t a 1920 x 1080 60p ccd?
Is it a 60i interlace chip?
Is use of the term “captured” in the above context have an engineering basis, I haven’t heard it used in this way before?”

“Captured” means here that the camera will produce 1920x1080p picture for further processing.
Single chips have lesser resolution, but they are aligned to each other by using spatial offset aka. pixel shift. We’ve been discussing about this with Jan (and thanks Jan for this incredible amount of attention we are getting) in another thread.

As for ccd’s being all progressive or some interlaced, I understand that usually when we talk about ccd’s we mean the whole camerahead consisting of the ccd’s and ad-converter and dsp after them.

I think that the ccd chip alone could be interlaced in a theory, because all ccd types (FF, FT & IT) have a serial register at the end where one horizontal line at a time is transferred. Ccd could be designed to tranfer only every other line to this register or the register could just disregard every other line. However, to my logics, it would be just simplier to build ccd to transfer the whole image to ad-converter and drop every other line in there or after that in dsp. Those things are easier to do in digital domain than in analog device like ccd.
Anyway all “broadcast quality” cameras, that have more horizontal lines in the ccd’s than cameras feed out, have to output all image data (progressive frame) to dsp, because there must be some scaling before interlacing. Even if ccd would give “readymade” interlaced picture out, it has to be low-pass filtered to avoid 30/25Hz flickering of the small details and to do this with high quality dsp has to compare both fields of the frame.

But I know this only in theory, so if there is somebody who is familiar with practical manufacturing of ccd’s, it would be nice to know more!

And, btw, does hvx have IT ccd’s?

[Toke]

Gotcha!
You guessed wrong!
I’m really not that young any more. 34 years, but I’m trying to stay youthful 😉
And I’m sorry if I’m too harsh and that offends someone.

But to be serious, when somebody says
“Without pixel shift/spatial offset, you would have vertical stripes in the pictures, which once the light became low you would see.”
, I need an eplanation.
I’ve been shooting for a decade with all sorts of cameras and almost none of them were using pixel shift and most of them give beatiful picture when used with skill. No vertical stripes everywhere.

Then either I didn’t understand what Jan explained of Jan wasn’t very clear how this pixel shift happens in a bottom technical level. I thought it would be good for all of us to understand this thing a bit better by having a discussion.

Maybe that was a bad idea…

[Toke]

[Jan Crittenden Livingston] “Scaling is taking a signal and telling a 960 X 1280 to become 1920 X 2560 with no additional information. It interpolates information to do so. This is not what happens at all, nothing is made up, it is taking information that is there and using it. That is why I am saying that it is not scaling.”

Ok, maybe this is again a lnaguage issue or we should define what these words mean.
https://en.wikipedia.org/wiki/Scaling
https://en.wikipedia.org/wiki/Interpolation

AFAIK, scaling goes both ways. It is resizing. Like with spatial offset camera has 2560×1440 image and it is scaled to 1920×1080.

Interpolation (AFAIK again), is calculating new samples between the old ones. So, in a way it is making new data out of nowhere. With spatial offset rgb-pixels are not on top of each other, so camera has to calculate the missing data, usually with averaging the neighbor pixel’s values (with no additional information).

Am I right or missing something, anyone?
Is this conversation boring or does not give any new or additional knowledge about camera technology?
I can stop this, if I want, really 🙂

[Toke]

[Jan Crittenden Livingston] “I have a feeling this is another one of those discussions that we will never come to agree as frankly it is not scaling. It is using information that is there and using all of it without making the pixels so small that it suffers in low light or has no dynamic range.”

You are telling why it is done, not how.

Scaling means altering pictures resolution.
Eg. when you have 2 ccd’s with 1280×720 resolution and v&h spatial offset, they will create a picture with resolution of 2560×1440. Then pixels are interpolated so that missing subpixels are filled and then picture is scaled down to final resolution.
How can you say that the last action is not done?

[Jan Crittenden Livingston] “But then you do nothave to be disappointed because it does use horizontal and vertical spatial offset.”

Ok, thank you for leaking this information. Japanise advertising was misleading by saying that it has 1080p-chips.

[Jan Crittenden Livingston] “Since the initial capture from the chip set is a 1080p/60 capture and it downconverts or crossconverts from there, there is not the limitation that you suggest here.”

I hope the initial capture can also be 1080p24/25/30, since the motion blur of combined 1080p60 frames cannot match the film cadence. You also get more light into the chips with longer exposure time and that means less noise from the chips.

[Toke]

[Jan Crittenden Livingston] “[toke lahti] “Isn’t it better that you get all color primary subsamples registered at the same spot, both in the ccd block and eg. in the hd-sdi feed?”

See this is where you keep thinking that the signal coming from the CCD is a dgitial signal.”

I’m not thinking that ccd is digital device. But analog data gets converted to digital and then it’s naturally better that all primaries color samples are taken from the same spot in the picture.

[Jan Crittenden Livingston] “Because this is not how it works. A spatial offset takes the red and blue chips and covers the non-photosensitive areas of the green chip. This allows a chip set to produce more resolution than what a single chip could do all on its own. These signals make an additive gain to the overall performance of the camera.”

I know what spatial offset is.
And it really doesn’t matter at what stage you do the ad-conversion. You can use spatial offset with 3ccd or 3cmos camerablocks.
Idea is based on a fact that analog rgb-data from ccd’s gets digitized and then converted to _component_ format, where luminance has more resolution than chrominance components.

You don’t get any more resolution with spatial offset to the rgb-stage, but since the luminance signal is calculated from all of these rgb-colors, you get the benefit in resolution after YCbCr-conversion.

So you do get some info from other chip’s non-sensitive area to produce a higher resolution luminance data.
Calculating this luminance data needs averaging rgb-data, interpolating and usually, yes, scaling.

[Jan Crittenden Livingston] “With spatial offset. It is not scaling, it is not cheating, it is engineering.”

This is where you go wrong. Spatial offset uses fundamentally scaling. It is same thing that happens in one-ccd/cmos digital still cameras. If camera uses offset in all 3 chips, then you get a raw picture which has double vert&horz resolution and all samples next to each other, side by side. From this image YCbCr-signal’s components are calculated with interpolation and scaling.

If you would like to get a rgb-picture as an end result, like with digital cinema cameras (viper/hdcam-sr), there would be no use for offsetting the chips, because you would just have to “onset” the samples back on top of each other. Moving samples (=pixels) spatially back and forth always blurs them a bit because all the roundings that has to be done in calculations. That’s why you’d want to get the rgb-pixels on top of each other at the first place, in the ccd’s.

And again there’s also no benefit for spatial offset with full resolution chips, because you already have full rez rgb-components, you can’t use more resolution to the YCbCr-components. Because 2/3″ ccd’s are usually full rez chips they don’t use spatiall offset.

Then again 1/3″ chips are so small that for them it is better to use non-full rez ccd’s to get larger area per pixel in the chip so the chip remains more sensitive. And that’s why I’m a bit dissapointed if hvx200 does not use vertical pixel shifting. My claim is that with 960×720 chips with vert&horz spatial offset, you could get as sharp pictures than with 960×1080 chips with only horz spatial offset. Previous just has more sensitivity.

Only logical explanation that I can come up on why no vert pixel shift is that it would be more difficult to use with interlaced picture. So interlacing, the great ghost of the analog age, still haunts us…

I think that as long as we are recoring in YCbCr-format, we could always benefit from spatial offset and it would be good idea to Panny to build next varicam with vert&horz pixel shifting 1280×720 chips to produce great 1080p-picture.

Maybe in the distant future, when everything has too much bandwidth and storage space all recoring change to rgb…

[Toke]

[Jan Crittenden Livingston] “And BTW, most CCD cameras that I am aware of use some form of offset technology. Most just don’t talk about it.”

Jan, could you enlighten me, why would full res ccd cam (like 2/3″ hdcam) use offset?

Isn’t it better that you get all color primary subsamples registered at the same spot, both in the ccd block and eg. in the hd-sdi feed?
Why would you displace these subpixels when there is no scaling?

[Toke]

[Ron Shook] “Jan said the camcorder goes into production in about a week.”

Actually Jan said last week, that camera goes to production next week, so right now there should be at least hundreds of ready made cameras. And that means that specs can’t change any more, so I just can’t think of any reason why still hide them. Is there anything to be ashamed?

[Toke]

[Lawrence Bansbach] “Jan has repeatedly said that the HVX’s CCDs will use pixel shift.”

Sorry, I don’t read DVXuser.
So there must horizontal pixel shift, because there is no vertical pixel shift, because the chips are “native progressive 1080/60p scan”.

[Lawrence Bansbach] “”Without pixel shift/spatial offset, you would have vertical stripes in the pictures, which once the light became low you would see.””

Could somebody explain what this means?
Every ccd and cmos have gaps between photosensitive area of the pixels.
This does not mean that those gaps are somehow sampled into produced picture.
If you like, you can think that the produced picture has “wider” pixels than the ccd, that overlaps the gaps.
So if camera would have full resolution chips (same amount of pixels on a chip than in produced picture), there’s no need or use or benefit from pixel shifting.

Since hvx200 does use horizontal pixel shifting and it produces a picture with max 1280 horizontal pixels, there must be fewer horizontal pixels on a ccd chip.
So, I guess it has 960×1080 chips, same as fx1/z1, so it also can’t be much more sensitive than those sony hdv’s.