Some quick, but technical HVX questions.

[Michael F. Grgurev]

Okay, been through some old threads, but still not quite sure where the facts stand on a couple things.

1. The HVX used both horizontal and vertical pixel shift.. yes? no?

<<< edit: Just stumbled upon it on my own.. the answers yes.

2. The HVX resolves a final video with a sampling equivalency somewhere between 4:2:0 and 4:2:2, recorded onto a 4:2:2 format (DVCPro on P2s)?

3. The HVX (for 720p) shifts it's 960 x 540 CCDs to 1280 x 720, then downconverts it (anmorphically squeezes?) down to a 4:3 960x720, due to the 4:3 aspect limitations of the DVCProHD codec itself... regardless of being on P2 or via firwire, etc?

<< edit: I've found the first part to be a bit off.. I realize. It's 1440 X 810 right? But in any case, is the rest accurate?

4. It's not as sharp (resolves less detail) as other HD cameras it's normally compared to, however it's likely better for screening and some other post-production activities... due to less compression, recording full actual frames, and being 4:2:2ish (getting better color accuracy (chroma to luma) then if it were true 4:2:0 on a 4:2:2 medium, but not as much as a true native 4:2:2.. or a 4:4:4 downsampled to a 4:2:2;... so to speak)?

.... and if the arguing's not over.. a simple "we can't agree yet" will suffice, though viewpoints are welcome. ;)


Oh, also... I'm a little shady on all the fine specifics of how pixel-shifting extrapolates detail.. though I know it's effectiveness varies depending on the color arrangements in the frame as well as with motion, apparently. But in any case, the question here is if anyone has found the use of filters (color/contrast/etc) to have any bearing on the shift effectiveness of a given shot (frame). I.E. to minimize, if mildly, the reduction of effectiveness in situation when the camera isn't shooting something that doesn't allow it to make as effective use of shifting? This is sorta a general question about pixel shift moreso then specifically the HVX200, I suppose, but it applied to the camera.. so yeah...

[Peter Jefferson]

Im not a genius on the camera, but i do know afew things about the formats in question..

2. The HVX resolves a final video with a sampling equivalency somewhere between 4:2:0 and 4:2:2, recorded onto a 4:2:2 format (DVCPro on P2s)?

((Not Equivalent.. it IS 4:2:2.
Closest thing to Digibeta your gonna get..
in an NTSC world, these differences are noticable, as are the differences between HDV 4:2:0 and DV 4:1:1 (refering to colour, NOT resolution).
However in Pal land, the colour sampling in DV AND in HDV is still 4:2:0..
There are no additional benefits for colour sampling in the HDV format vs PAl (again.. colour not res or codec)...not in PAL land anyway.. in fact.. there is a compromise considering the pixel size of the HDV CCd's in question, and in most cases, dynamic range of HDV acqured footage vs SD D/Range isnt as high/wide within HDV.. this isnt the codec... its the amount of light which hits the CCD...all this plays a major part in the colour reproduction))

3. The HVX (for 720p) shifts it's 960 x 540 CCDs to 1280 x 720, then downconverts it (anmorphically squeezes?) down to a 4:3 960x720, due to the 4:3 aspect limitations of the DVCProHD codec itself... regardless of being on P2 or via firwire, etc?

((err... from my understanding, its scales it to the appropriate resolution... however the footage is either captured at native 16:9, or native 4:3
1280x720 or 1440x1080... be it 4:3 or 16:9... aspect ratios bears no weight on the codec.. not from my understanding of the codec itself. ))

<< edit: I've found the first part to be a bit off.. I realize. It's 1440 X 810 right? But in any case, is the rest accurate?


((not that i know of mate.. i dont know where u got ur numbers from but the files themselves are or should be using the appropriate "standard" HD resolutions.. even if they arent resolved natively.. then again, tehr only camera that does shoot native HD resolutions is the JVC.. the Sony and canons all upscale as well... albeit in their own special way... canons technique seems to be the one to offer the sharpest reuslt in this "scaling" ))

4. It's not as sharp (resolves less detail) as other HD cameras it's normally compared to, however it's likely better for screening and some other post-production activities...

((Not likely better... IS better... this is due to bitrates, colour sampling and codec. In addition the HVX has the highest dynamic range of all the HD cameras within this range of units. This is due to the larger pixels <= lower res> within the CCD, as well as the Gamma configurations within the camera itself))

due to less compression, recording full actual frames, and being 4:2:2ish (getting better color accuracy (chroma to luma) then if it were true 4:2:0 on a 4:2:2 medium, but not as much as a true native 4:2:2..

((well it IS true native 4:2:2 , so i dunno where ur gettin ur info from about htis "whereabouts" or "close to"...
However the reasons you mentioend are predominately the reasoning why its such a good keying unit.. ))

or a 4:4:4 downsampled to a 4:2:2;... so to speak)?

((dont get ya.. slow down... take a breath bro... ))

.... and if the arguing's not over.. a simple "we can't agree yet" will suffice, though viewpoints are welcome. ;)

((What argument? Every camera has its uses.. and if people cant fathom that, they need to get a life... there are more important things to argue about.. let alone argue about a camera which does what it says it does))

Oh, also... I'm a little shady on all the fine specifics of how pixel-shifting extrapolates detail..
((extrapolate, or interpolate?? ))

though I know it's effectiveness varies depending on the color arrangements in the frame as well as with motion, apparently. But in any case, the question here is if anyone has found the use of filters (color/contrast/etc) to have any bearing on the shift effectiveness of a given shot (frame).

((err.. no.. the shot is the shot.. and it is captured as is, then scaled to fit within the given HD resolution. What you do with it after that, bears no weight on how pixel shifting works... <hang on a bit, theres more.. >

In camera, the imaging and "tweaking" is done prior to the hard copy being printed into tape/P2... again, as the processing is done PRIOR to recording, it shouldnt make a difference whatsoever.. <i know u didnt ask that...> BUT the compression may be an issue.. thats one point.. coz what u see on screen is NOT compressed...

All pixel shifting does (in this cae) is scale what we see per pixel... much like taking a 3mp still with a MX500... the actuall CCD is 880k pixels <less than 1 mp>, but when it interpolates to give u a 3mp picture, all its doing is scaling.. nothing more. Much like the way an NLE woul dscale SD footage up to HDV

In addition, pixel shifting allows colour to be defined between varying contrasts.. or better known as edging... to differentiate betwen colour variable or variable shades.

This can be considered artificial sharpening even.. and to the najed eye. it is..
As some shifting technogies sharpen the image (as found in the DVX100...) and finally when tweaking the sharpness, the pixel shifting is the first to be affected by this.. as sharpening is shaving off the gradation between colours withn the fine lines of one colour to another... Lets say a talking head with a guy in a blue shirt... ove rhis shoulder we see a bookshelf of varying colours.. when u sharpen, your defining the edge line between that final BLUE point of his shirt, the to edge point of the bookshelf.. sometimes this causes jaggies other times it causes ghosting.. depnds on the camera.. sharpenign and pixel shifting are two differnt things however sharpening affects how shifting works...
Contrast and saturation and gamma also play a part in how shifting works, BUT not on the same level (as those "edge lines" between colours are what the pixel shifting engine will look at when it comes time to make the final image).
In most cases those edge lines bleed.. DV moreso than other codecs, due to the low bitrate and aspect ratio (as in pixel aspect) HDV is square, and from what i know of DVCProHD, it too is square in its native form. DV50 and 25 are NOT square they run either 16:9 or 4:3

another issue is red saturation, however in DVCPro50/HD this should be too much of a problem. unless u crank it up in the scene files of course.. then u might get bleeding within the pixel shifted areas.. ))

I.E. to minimize, if mildly, the reduction of effectiveness in situation when the camera isn't shooting something that doesn't allow it to make as effective use of shifting?
((Huh?? shifting is ALWAYS happening.. whether it be simple talking head stuff or v8 supercars where pans are stupidly fast... it doesnt matter.))

hope this helps.

[Barry Green]

Originally Posted by Michael F. Grgurev

2. The HVX resolves a final video with a sampling equivalency somewhere between 4:2:0 and 4:2:2, recorded onto a 4:2:2 format (DVCPro on P2s)?

The recording format is 4:2:2. The footage it delivers is far and away the best for chroma keying; according to some guys from WETA Digital it delivers better keying than an XDCAM-HD system. As a recorded format it's 4:2:2. The chips may not be capable of delivering a full 1920x1080 worth of data, but they do deliver more than enough that the downsampled recorded image is 4:2:2.

3. The HVX (for 720p) shifts it's 960 x 540 CCDs to 1280 x 720, then downconverts it (anmorphically squeezes?) down to a 4:3 960x720, due to the 4:3 aspect limitations of the DVCProHD codec itself... regardless of being on P2 or via firwire, etc?

Well, no. I mean, sort of, but not.

The chips are sampled into a 1920x1080 matrix. All internal processing is done at 1920x1080. However, they don't get a full 1920x1080's worth of detail out of it; there's only so much detail that can be extracted off 1/3" HD chips and even then some concessions were made to noise/dynamic range/etc instead of ultimate maximum detail. So it gets about 1440x810's worth of sharpness out of its chips, processed within a 1920x1080 matrix internally.

Then that 1920x1080 gets scaled to whatever the recording format is. So for US it's 1280x1080 for 1080i/p, for EU it's 1440x1080 for 1080i/p; for 720 it's 960x720, and for SD it's either 720x576 (EU) or 720x480 (US). But it all starts from that original 1920x1080 scan.

About that 4:3 comment -- there is no "4:3 limitation" of the codec (else how does one explain 1280x1080?). 960x720 was chosen because it's exactly twice as many pixels as the standard-def DV codec. 720x(480x2) = 720x960. They use two codecs ganged together for the frame size, and two more codecs ganged together to support the frame rate, and the result is 4x as much data (100 megabits instead of 25 megabits for a single DV codec). They standardized on the 720 because vertical resolution is what's most important, so they got full vertical resolution, and then for horizontal they got as much as was possible, as dictated by the bandwidth of the codecs. Which is why 1080i/p mode is 1280x1080; they started with 1080 and used as much width as they can get. 1280 x 1080 x 29.97 = the exact same number of pixels as 960 x 720 x 59.94.

4. It's not as sharp (resolves less detail) as other HD cameras it's normally compared to

Here I'd absolutely disagree. It's every bit as sharp as the Sony and JVC. None of them are as sharp as the Canon XLH1 though.

, however it's likely better for screening and some other post-production activities... due to less compression, recording full actual frames, and being 4:2:2ish (getting better color accuracy (chroma to luma) then if it were true 4:2:0 on a 4:2:2 medium, but not as much as a true native 4:2:2.. or a 4:4:4 downsampled to a 4:2:2;... so to speak)?

The rest of it I'd basically agree with, except that I'd say that the recorded image is a full 4:2:2.

Oh, also... I'm a little shady on all the fine specifics of how pixel-shifting extrapolates detail..

No "pixels" get "shifted". Pixel Shift is more appropriately called spatial offset, because one of the CCDs is offset from the others. People who don't understand anything about the technology think of it as RGB up-rezzing, which it isn't. The process is more like a demosaic on a bayer-filtered image. Look at the RED camera; it's a single-chip camera that delivers its color image through a demosaic process. Spatial offset works somewhat similarly. Each of the three chips (R, G, and B) is sampled to deliver the YUV components; a pixel's Y value (luminance) is derived from a formula that's something on the order of 60% green, 29% red, 11% blue. All three chips are used together to determine the luminance value of each pixel.

Exact technical breakdowns of the process are exceptionally sparse on the web; Panasonic published a tech paper to describe how the CCDs work, and the BBC wrote a white paper where the technology was briefly mentioned; they said that it holds the potential of delivering 1.5x as much resolution but as a practical matter the maximum was more like the square root of 2 (1.414x as much) and as an observable increase, 1.33x was a reasonable factor.

[Michael F. Grgurev]

Thanks for the replies.

In regards to the 4:2:2, I stated what I've read elsewhere regarding the color sampling when pixel-shift. Although I didn't really take it in much faith first, after some heavy thought into it, it does make some sense to me how at the very least the luma to chroma relationship would be a bit different then on a similar CCD that doesn't apply the shifting method. That simply being because your effectively increasing the luma resolution, but not the chroma. Basically color sampling terminology list 4:4:4, 4:2:0, etc.. as I understood them anyway, are all about ratios, and in such a case you would be effectively changing the ratios between the Luma and Chroma resolutions. I'm not necessarily saying it has a huge impact.. just that it'd be kinda different.. depending on whatever method cameras use to internally process all that data.

For instance, from my limited understanding, it seems to me that if you were to raise the luma resolution in one direction.. by pixel shift or whatever, and you were -hypothetically- able to raise it by 200%, then the actual color sampling ratio would be 4:2:2 exactly... because you have twice as much luma info going one way as chromo info. Furthermore, if you do the same on another axis (by our -hypothetical- 200%), then it'd only come to a 4:2:0 color sampling. That being said, your NOT losing color info, your just gaining luminance info (resolution), and 4:2:whatever is just a ratio between the two.

Originally Posted by Barry Green

No "pixels" get "shifted". Pixel Shift is more appropriately called spatial offset, because one of the CCDs is offset from the others.

Heh, well it's sorta depends on you look at it. Technically each photoactive region on a the green CCD, each corresponding to pixel, is "shifted" (physically) by %50 a region width, in one or two directions. The resulting video, in RGB state, has a green channel with detail shifted in relation to the other channels. But I get what your getting at. ;)

Originally Posted by Barry Green

About that 4:3 comment -- there is no "4:3 limitation" of the codec (else how does one explain 1280x1080?).

Herm, I guess I might be FUBARing terminology here, I suppose. Or maybe it has something to do with pixel shape?.. though that doesn't seem right

Basically I was referring to the actual ratio (aspect ratio) of anything at those resolutions.. (and being square pixels in a physical world). Like, for instance.. 1280x1080... plug that into Photoshop... it's a 4:3 aspect ratio. The actual HDTV (widescreen) standard for 1080 is "1920 x 1080". For instance, like you said.. it's 4x as much as DV (StandardDef), then that's 2x in each dimension thus still effectively the same aspect ration.

So yeah, I realize I must be missing something... are we talking about cross comparison between square (actual) and non-square pixels?

[Barry Green]

Originally Posted by Michael F. Grgurev

Basically I was referring to the actual ratio (aspect ratio) of anything at those resolutions.. (and being square pixels in a physical world). Like, for instance.. 1280x1080... plug that into Photoshop... it's a 4:3 aspect ratio.

How?

4:3 in photoshop would require 33% more pixels horizontally than vertically. 640x480 is 4:3. 1280x960 is 4:3. 1280x1080 is not. It's 1.185:1.

The actual HDTV (widescreen) standard for 1080 is "1920 x 1080".

Yes, but no common recording format records at 1920x1080, they all prefilter to a smaller dimension, then use a pixel aspect ratio on anamorphic pixels to get back to the desired end product of 1920x1080. So HDCAM and HDV and 50Hz DVCPRO-HD all use 1440x1080 with a PAR of 1.333, 60Hz DVCPRO-HD uses 1280x1080 with a PAR of 1.500.

So yeah, I realize I must be missing something... are we talking about cross comparison between square (actual) and non-square pixels?

I don't know. There is no 4:3 "ratio" in 1080 mode, nor is there a 4:3 ratio in standard-def mode in either PAL or NTSC. The only mode that mathematically works out to 4:3 is 720p mode, and I suspect that's more just raw coincidence than anything planned on by the designers.

[Michael F. Grgurev]

Originally Posted by Barry Green

How?

4:3 in photoshop would require 33% more pixels horizontally than vertically. 640x480 is 4:3. 1280x960 is 4:3. 1280x1080 is not. It's 1.185:1.

Herm... I see where I muffed up with my explanation.. but I think I'm still sorta right.. or we both could be actually. Took me a second to get my head around it for some reason. Basically, it all depends on what you mean by 4:3. 4:3 taken as a general ole' ratio, yea... would seem to apply to 640x480. However, I was thinking of the effective output for television when I was talking about 4:3, the visual aspect ratio. I misstated that whole "square pixels in a physical world" bit.

Essentially, 640x480 can be expressed as a ratio 4:3 (horizontal resolution by vertical resolution), but it could be somewhat wrong to say a frame/output of 640x480 worth of actual resolution, would be displayed as a 4:3 aspect ratio image on a TV. I mean it could be I guess, stretching and whatnot, but from a technical standpoint, NTSC utilizes non-square pixels, with a 0.9something width. In other words, that's why a 720x480 image can be said to be 4:3 on a TV. (720x0.9) x 480.

Originally Posted by Barry Green

I don't know. There is no 4:3 "ratio" in 1080 mode, nor is there a 4:3 ratio in standard-def mode in either PAL or NTSC. The only mode that mathematically works out to 4:3 is 720p mode, and I suspect that's more just raw coincidence than anything planned on by the designers.

So yeah... I suppose it depends on what we're talking about... video on a TV or a computer display or whatever. I kinda figured terms like "4:3" to mean what the actual display ratios would be on the medium it was intended for..in this case I mean NTSC television.. lets say SD for the time being. In which case there is a 4:3 ratio in standard def. Matter of fact, most standard def video is 4:3.. it's the standard. The fact that there is no 4:3 ratio in 1080 mode general is true in all cases. 720p is 4:3 at 960x720.. because HDTV uses square pixels.. as I understand it. So the 16:9 (on an HDTV) 1280x720 picture is then anamorphically squeezed into a 4:3 960x720 space when on tape or P2 or wherever.. with DVCPro HD. Basically the same thing MiniDV does with widescreen.. when it's done the better way.. not by letterboxing and wasting storage space. 1080i/p is the exception.. I think.. it's sorta weird. DVCPro HD only records to 4:3 in PAL... in NTSC it's 1280 x 1080 (60Hz vs 50Hz). I guess it's bandwidth limitation, as you said... and the fact it's interlaced.. since there really is no 1080p.. so the specification is weird for PAL vs NTSC.

So yeah.. my original point was just that DVCProHD records 720p to tape (or P2/drive) anamorphically.. at what could be considered a 4:3 aspect.. just like MiniDV records widescreen. Which I'm not saying is bad or anything.. just saying that's what everything does. Heck.. doesn't film work more or less similarly as well? 1080i is the odd-ball which I don't know much about.. but it still records onto a virtual 4:3 region.. in PAL.. and NTSC it just seems doesn't use the full horizontal space... as -if- it were slighty letterboxed almost.