If you want to test a format, you can import a few seconds of the static or moving 4:4:4 chart to make a chort clip and export to the codec of your choice. But remember that codecs vary in quality. One computer DV codec can be better than another and a camera codec is again different.
Finally! CCD specs and DSP released!
- Thread starter Jarred Land
- Start date
BobDiaz
The Professor
Yuval Shrem said:
IF we are talking about SD (DV25 or DVCPRO50) footage, there are more than enough color pixels to do a greenscreen OR bluescreen and it will work great either way.
As for 720p footage, in the VERTICAL direction we have 75% of the color pixels necessary for either a good bluescreen or a good greenscreen. In blowing up images, I can see a clean dividing line between the greenscreen and the person. You could say that at 720p, the camera acts like 75% 4:2:2 and 25% 4:2:0. This is better than Sony, JVC, or Canon's 4:2:0 color. In the horizontal direction, Panasonic's 960 color pixels are more than enough.
It is the 1080 footage that the pixel count becomes an issue. Let me define terms:
TRUE 4:2:2, in this mode every new scan line receives new RGB color information.
OFFSET COLOR 4:2:2 (or Offset Color 4:2:0), the HVX-200 receives new color information on every scan line, but it is either RED+BLUE or GREEN. In 1080 mode, this sits half way between the TRUE 4:2:2 and the 4:2:0. So, it's better than 4:2:0, but not as good as TRUE 4:2:2. In the horizontal direction, the HVX has enough color pixels to do a proper horizontal 4:2:2.
4:2:0, This is how HDV records on the Canon, Sony, and the JVC. Every 2 scan lines the Color information changes. (Anyone, correct me if I'm wrong, but this is how I understand 4:2:0 works.)
In the 1080 mode, a greenscreen is the best choice, because the other color is picked up one scan line down. With bluescreen, if the other color is green, the green is picked up one scan line down; so the HVX still acts like OFFSET COLOR 4:2:2. The red is on the same scan line as the blue, so with reds the HVX acts like 4:2:0. You can still do a key, but it is not as good as 4:2:2.OFFSET COLOR 4:2:2 (or Offset Color 4:2:0), the HVX-200 receives new color information on every scan line, but it is either RED+BLUE or GREEN. In 1080 mode, this sits half way between the TRUE 4:2:2 and the 4:2:0. So, it's better than 4:2:0, but not as good as TRUE 4:2:2. In the horizontal direction, the HVX has enough color pixels to do a proper horizontal 4:2:2.
4:2:0, This is how HDV records on the Canon, Sony, and the JVC. Every 2 scan lines the Color information changes. (Anyone, correct me if I'm wrong, but this is how I understand 4:2:0 works.)
With the Canon, you do NOT gain any more color resolution in the vertical direction. It records 4:2:0, which gives 540 color changes. As for resolution, while the camera is 1080, we've seen that the resolution gain is about 12% more equivelent pixels than the JVC. Yes, you record more pixels than the JVC, BUT you do NOT record that much more detail.
I have still images of greenscreen shots, but don't have any of a bluescreen shot. So, I have not had a chance to see how it looks in 720p or 1080i mode. So, I have to work from theory rather than images. (If anyone wants to post a still bluescreen image at 720p and 1080i, I'd love to see it.)
Currently I've generated 10x cuts from different greenscreen shots so everyone can see every pixel. DV25 is poor, DVCPRO looks great, 720p is really good, and 1080i has funny glitchs in it and I don't understand why. Canon's Greenscreen clip has 4:2:0 glitches in it, so don't panic. School comes first, so it may have to wait until Friday before I can write this up and post the images.
Bob Diaz
Offset color does not exist as a term. I'm surpised to hear that from someone who teaches others. A chroma position must contain all three colors. Expand the area untill you hit at least one R one G and one B PHYSICAL PIXELS. That is your chroma unit. Measure this area after converting to 1280x1080. It is 2x2pixels. I can smooth a single huge square of chroma to fill the entire screen and I will still have "transitional pixels" from that single pixel. We are talking about 540pixels here. Nothing can be done to increase that. You can smooth it but it still 540pixels which is not 4:2:2 in any way, true or false or semi or whatever. It is 540 VERTICAL PHYSICAL PIXELS and that's it.
btw the drawings on the article have the RED and BLUE sensors without shift between them. It would be a silly thing to shift those since that would only create a color problem that would have to be smoothed. Nothing to gain and Panasonic chose right. The luma gain is already there from the green sensor shift, so why compromise color with no gain at all?
btw the drawings on the article have the RED and BLUE sensors without shift between them. It would be a silly thing to shift those since that would only create a color problem that would have to be smoothed. Nothing to gain and Panasonic chose right. The luma gain is already there from the green sensor shift, so why compromise color with no gain at all?
Last edited:
BobDiaz
The Professor
That's true, but we are no longer talking about shooting to HDV. In the context of what I was writing about, 4:2:0 is what HDV uses. Now, IF someone wants to take the component out of the Canon, depending one the system chosen, it could be 4:2:2, but as far as recording to tape, it's 4:2:0.
As a standard term, you are correct, but I know of no standard term to describe what Panasonic is doing with the HVX. My failure was to make it clear that the term I defined, was only in the context of my posting and not a standard term everyone uses.
Bob Diaz
ullanta
Veteran
BobDiaz said:
No failure, it was pretty clear. Limo, you have to open your mind just a bit. A chroma position indeed has to have all three colors, but each pixel may not be exclusive to that position. What Bob is sensibly saying is that in the Panasonic case, we may describe a chroma position as a unique COMBINATION of R,G,B pixels. There may be some overlap, but each TRIO is unique and has some new color info. Doesn't fit into standard terms, perhaps, but makes sense.
ullanta said:
No. L3 (aka L-cubed) stands for "Layered, Low-complexity, Lossless," with a 2.5 to 1 compression ratio. Some of the 8-million pixel (4K x 2K) cameras are using it. It crunches down to 1.2 Gb/sec (reportedly ten times the data stream of HD 4:4:4 10-bit). It's supposed to be completely lossless, and because it's a low-complexity algorithm, you don't need complex hardware to run it. Sounds like an efficient and effective technique.
johncordell
Member
ullanta said:
I can see how pixel shift yeilds more luma info, that makes sense to me. But I don't quite see how it yeilds more chroma info. For instance, if you didn't actually shift the green, but still did the exact same calculation as the HVX does, you'd have 'unique' RGB combinations in each of the generated chroma samples too. But in that case it would be called a form of interpolation, not added chroma info, right?
Once you do shift the green, if indeed there's a chroma benefit to pixel shifting, something good must happen. Seems like it would relate to the spatial accuracy of each generated chroma sample as compared to the non-pixel shifted case. Doesn't seem impossible to me that it would create a benefit, but I don't see exactly how and by what measure.
The 'uniqueness' of the triplets alone doesn't convince me of the added chroma information, if that was the case I could go grab random R, Gs and Bs from all over the map and create new and wonderful unique RGBs.
The percentage benefit to luma of pixel shifting seems well understood and exactly measured. I've not seen any equivalent treatment describing how pixel shifting improves chroma resolution. If anyone has, I'd love a pointer to it. My intuition tells me it would be 'diffferent' but not mathematically better. Like maybe certain color patterns might do better under one scheme or the other. Who knows, maybe it is just better to the eye for the kinds of things humans like to look at. Then again maybe not.
Oh, I should point out I don't actually know much about this subject, just applying common sense and a little arithmetic and not quite seeing how the chroma story in pixel shifting adds up.
toke lahti
Well-known member
If you compare the amount of data, they are equivalent.Barry_Green said:
If you compare them to human vision 4:1:1 has no sense at all.
Just a relic from interlaced era.
I see what's being said, and it's an interesting point. Engineering wise I'd think it to be a good way of doing things.ullanta said:
But all I think it will do in practice is smooth out the changes, very comparable to increasing the image resolution of a picture in a photo editing program. That will also seem to look better because it's less 'steppy' - but does not, can not, increase actual resolution. Same here.
Is this relevant to the real world? Only noticeably in certain cases probably, and chroma key being the most obvious, I'd say. Imagine a black horizontal bar against the green screen background, such that if you read out the values of green pixels 2=full signal, 0=no signal. If we read out a vertical column of the 540 green pixels, we may get: 2,2,0,0,2,2 etc. (Guess where the bar is!?!)
If we wanted 1080 pixels in total, we could simply repeat each one, so: 2,2,2,2,0,0,0,0,2,2,2,2. What Bob is saying (which I think is what does happen) is that the interpolation would give us something like 2,2,2,1,0,0,1,2,2,2,etc - smoothed, but no extra detail. Whereas 1080 real pixels may pick up a crack in the bar and give 2,2,2,0,2,0,0,0,2,2,2,etc say. Unlike the interpolation, they have found real extra detail.
Coming back to the real world, Panasonics approach I consider in principle to be a very good engineering compromise in designing a 1/3" HD camera. What I don't believe it can do is deliver more than 540 lines of real vertical colour resolution. If anybody can think of a mechanism whereby it can, I'd be very interested to hear it.
Last edited:
Ernest_Acosta
Well-known member
Now I think an HVX Andromedized will be worth the price. Juan we are patiently waiting.
BobDiaz
The Professor
I wanted to try to help clear up some of the things I said in the thread. This may get a bit technical... so be forewarned.
When I spoke about 4:2:0, I was speaking about the USA form of 4:2:0, where the color information changes every 2 scan lines.
With 1080 lines, a 960x540 CCD must have every other scan line as interpolated. This does NOT add any new information to the image, but it does smooth any rough edge in the color information. In 1080 mode, it looks like this:
In zooming into different still images of HVX in 1080 mode, I can't find any color problems due to their native CCD resolution and DSP. The exception is one green screen shot, which had glitches due to a codec problem. The problem was the conversion to another format and NOT the fault of the HVX.
You can see several 10x stills from different images in the following message:
http://www.dvxuser.com/V6/showthread.php?t=47914
Bob Diaz
When I spoke about 4:2:0, I was speaking about the USA form of 4:2:0, where the color information changes every 2 scan lines.
With 1080 lines, a 960x540 CCD must have every other scan line as interpolated. This does NOT add any new information to the image, but it does smooth any rough edge in the color information. In 1080 mode, it looks like this:
...
Scan Line 100: New Green, Interpolated Red & Blue
Scan Line 101: Interpolated Green, New Red & Blue
Scan Line 102: New Green, Interpolated Red & Blue
Scan Line 103: Interpolated Green, New Red & Blue
Scan Line 104: New Green, Interpolated Red & Blue
Scan Line 105: Interpolated Green, New Red & Blue
...
This is NOT as good as having new Green, Red, and Blue information on every scan line (true 4:2:2), but is better than repeating the same color information for 2 scan lines (USA 4:2:0).Scan Line 100: New Green, Interpolated Red & Blue
Scan Line 101: Interpolated Green, New Red & Blue
Scan Line 102: New Green, Interpolated Red & Blue
Scan Line 103: Interpolated Green, New Red & Blue
Scan Line 104: New Green, Interpolated Red & Blue
Scan Line 105: Interpolated Green, New Red & Blue
...
In zooming into different still images of HVX in 1080 mode, I can't find any color problems due to their native CCD resolution and DSP. The exception is one green screen shot, which had glitches due to a codec problem. The problem was the conversion to another format and NOT the fault of the HVX.
You can see several 10x stills from different images in the following message:
http://www.dvxuser.com/V6/showthread.php?t=47914
Bob Diaz
johncordell
Member
BobDiaz said:
But isn't a more fair comparison to have the missing lines in the 4:2:0 case be interpolated instead of repeated? If interpolation makes sense in the pixel shifted case, wouldn't you use it in the non-pixel shifted case as well? Agree the offsetting of the green might result in smoother interpolation, but the same *amount* of interpolation is happening in both cases, isn't it? I am curious as to how quantifyable the resulting chroma benefit is.
Thanks for the pointer to the green screen thread -- interesting.
All fully agreed.BobDiaz said:
Well - only if you're using the information directly as such. But surely if the keyer or whatever makes use of the 4:2:0 data via interpolation (as it should), it all comes out the same? Panasonics approach makes life simpler later in the chain, as it interpolates early on, then records the interpolated samples. The disadvantage to this approach is that bitrate gets used up recording interpolated samples - surely (in this idealised discussion) it's far more efficient in bandwidth terms to record 4:2:0 and then interpolate, rather than interpolate and then record 4:2:2?
BobDiaz
The Professor
As I tried to find images to prove or disprove my point, that the sharp edges of objects in 1/3" CCD images do not exist. Sorry, but 1/3" CCDs just do not provide full 1080 detail. What I see in the 10x views from the Canon and Panasonic are slightly fuzzy transistions from one edge to another.
In the case of the Panasonic, the transistion is smooth. In the case of the Canon, it can be rough, BUT yes there is software that can smooth out the color information making it a smooth transistion. Would it be as good as the Panasonic? I can't be sure until I see results.
Here is a 10x zoom green screen shot from the Canon:
Here is a shot from the HVX that I've pulled a 10x zoom from the right side of the face with something green in the background.
Here's the whole image in context; I reduced the size of the image to fit nicely within this message.
As seen cropped, look closely at the edge this is shown 1x size:
The background is a bit out of focus, but the edge of the face looks clean without any roughness to it.
I wish we had 2 exact same shots, one with the HVX green screen and the other a Canon green screen. Might be interesting to compare the final keyed results to see if there's any real difference.
http://www.coax.tv/ListOfIssues.htm#larrythorpe
While he does not get into color issues, he does talk about resolution.
Bob Diaz
In the case of the Panasonic, the transistion is smooth. In the case of the Canon, it can be rough, BUT yes there is software that can smooth out the color information making it a smooth transistion. Would it be as good as the Panasonic? I can't be sure until I see results.
Here is a 10x zoom green screen shot from the Canon:
Here is a shot from the HVX that I've pulled a 10x zoom from the right side of the face with something green in the background.
Here's the whole image in context; I reduced the size of the image to fit nicely within this message.
As seen cropped, look closely at the edge this is shown 1x size:
The background is a bit out of focus, but the edge of the face looks clean without any roughness to it.
I wish we had 2 exact same shots, one with the HVX green screen and the other a Canon green screen. Might be interesting to compare the final keyed results to see if there's any real difference.
Real World Tests ---> :nads: <--- Theory
You may find Part 3 interesting to read; see below:
http://www.coax.tv/ListOfIssues.htm#larrythorpe
While he does not get into color issues, he does talk about resolution.
Bob Diaz
toke lahti
Well-known member
With 4:2:0 you only have one chroma component sample per 2x2 pixels. There's no room for interpolated sample.johncordell said:
If 4:2:0 is converted to 4:4:4 (eg for cc or fx) then its up to decoder how well it interpolates the missing values.
Pixel shifting is two-bladed sword. On the other hand we get more luma resolution (but with lesser MTF) with real world images with all the colors in them. But in the trade off we loose chroma resolution in special situations like chroma key.
More calculated luma samples, but all a bit blurry and all calculated chroma pixels a bit blurry. Because every YCbCr pixel has interpolation in it.
Think about shooting a black object with green background. Or red, blue and black. Suddenly pixel shift advantage is all gone and all you have is 960x540 from which you have to substract Kell factor so you have only about 750x400. This is way lower than what chroma resolution camera records (1080p: 640/720x1080 720p: 480x720). Actually chroma resolutions that camera head produces and storage part records are upside down. So in the end there will be only 640/720x400 (1080p) or 480x400 (720p) in these cases. So we are quite far from what real fullHD 4:2:2 chroma would be (960x1080).
Once again I have to say that 4:2:0 would be more optimal chroma resolution for this camera when shooting progressive with square pixel luma resolution (1280x720).
Another thing with pixel shift is decreased "dynamic resolution" when there is movement in the picture. After all we are shooting moving pictures. But I don't think that's such a bad issue because human eye's resolving power also decreases with movement and there's also motion blur involved with moving objects.
Could somebody explain why this "dynamic resolution" decreases with pixel shift?
I haven't fully understood this yet.
toke lahti
Well-known member
Please, somebody, educate me?toke lahti said:
I suppose Steve Mullen does not chat here?