How come no HD cameras deliver anywhere near the horizontal resolution they should?

DavidBeier

DavidBeier

Veteran
Ok, now I'm aware that the eye is less sensative to horizontal resolution. I'm also aware that because of this the cameras often subsample the horizontal resolution. That said, I'm still confused why the the various HD cameras, even some of the best, all seem to display less horizontal resolution than they should.

Lets start with the HVX200. It's 960x540 chips pixel shifted to 1440x810, uprezzed to 1920x1080 and then subsampled at 1280x1080. Now the resolution tests have greatly varied. The best results I've seen is 670x730 and the worst were Adam Wilt's claim of 540x550. Forget about disputing which set of numbers is right, either 670 or 540 is still not what a 960 CCD should deliver.

Or lets look at the JVC HD100. It has 1280x720p chips sampled at the same resolution. Wilt claims he saw 700x700 lines of resolution (though some have disputed this). Even if that's true, isn't it odd that it gets near it's full vertical resolution but not near the full horizontal?

Or the Varicam. 1280x720 chips sampled at 960x720p. Wilt claims greater than 700 lines in both directions but shouldn't we be able to get a full 960 in horizontal? Especially from such a high quality model.

Not even the top of the line Cine Alta seems to be delivering. With 1920x1080 CCDs sampled at 1440x1080, the reported 1200x1000 lines is impressive but once again why isn't the horizontal number comparing with the vertical. Why aren't we managing 1400 lines?

Can someone explain to me what I'm missing here? I mean, we wanted true 16x9 CCDs for so long but it seems that most of the cameras aren't even delivering 4x3.
 
it sprobably the Data rate. either you compress more or fully represent less data. i suppose Fully represented less data, is "cleaner".
 
Nothing to do with the data rate, as when we shot those cameras we were monitoring the live head output, uncompressed.

David, the problem with trying to understand this is: it doesn't work that way. This is the same type of problems we run into when trying to explain that the # of pixels on the CCD do not necessarily equate to a direct understanding of the sharpness conveyed. People think there's a direct relationship, and there isn't.

You should look at Adam's article on DV.com about aliasing.

For example, let's take your desire that a 1920x1080 CCD should be able to deliver a resolution of 1920 horizontal and 1080 vertical lines, right? How could you possibly measure that? One way would be to print out a chart that was exactly, perfectly, a checkerboard or grid pattern of 1,920 x 1,080 cells. If you aimed the 1920x1080 sensor at that checkerboard, and you exactly perfectly lined everything up, then yes -- you should see the imaging head rendering exactly that 1920x1080 grid.

But what happens if you're a micron off to the left or the right? Or up or down? Well, see, now those boxes won't line up with the pixels on the sensor, will they? The nice clean black lines that you've been seeing will now instead hit BETWEEN two pixels. So each pixel on the CCD is seeing a chunk of black line and a chunk of white line. What is that single pixel to do? It of course renders gray -- what else can it do? So now your nicely resolved 1920x1080 frame, where you were seeing 1920 lines of horizontal resolution -- that same camera, rendering that same frame, is now going to show a sheet of solid gray. Zero lines of resolution.

So which is it? 1920 lines? Or zero? The obvious answer is: it's neither.

A 1920x1080 sensor could never deliver 1080 full vertical lines, or 1920 horizontal lines, except in the 1-in-1000 case where the lines happen to exactly perfectly line up with the pixels in the center. In any other case it's going to involve shades of gray.

What happens if we remove the camera entirely from the equation? I took a high-res PDF of an ISO 12233 res chart and blew it up to a huge size, and trimmed it to exactly the 16:9 frame. Then I used PhotoShop to resize that down to 1920x1080. This should be a best-case scenario, no camera or lens or optics in the way to mess anything up, no aliasing, etc. I then viewed it at 100% magnification on my 1920x1200 monitor so the full image is shown at native pixel res. And it looks like it's resolving about 1000 x 900.

That's without horizontal prefiltering. If I shrink that image down to 1440x1080 and then expand it back up to 1920x1080, the resolved image changes to about 900 x 900. If I instead shrink it from 1920 to 1280 and then back up, it looks like it's resolving about 830x900.

That's with no camera involved, no HDV or DVCPRO-HD encoding going on, that's just how the black and white bits line up.

So you're not going to see 1920 horizontal lines from any of these cameras (except maybe the Red, where with 4x as many pixels, the black and white lines can be discerned to a finer degree). But even then, you have to realize that it's not that simple (it never is). Red doesn't use its pixels as-is to make an image; it uses a bayer pattern filter and has to de-mosaic its image. So each pixel's value in the recorded frame is not made from a corresponding pixel on the sensor, but instead (to put it simply) it depends on the values of multiple pixels on the sensor. It's sort of like pixel shift in that way. You don't read out a sensor and map its pixels directly to the image. That could only work in black and white (keep in mind that CMOS and CCD sensors are black and white devices!) To get color you typically either use a 3-chip system, and use information from all three chips to determine a pixel's color and brightness value, or you use one sensor and a bayer pattern filter.

So there's no 1:1 relationship. Even if there WERE a 1:1 relationship, you can't get at the pixels that way, as CCDs are analog devices. Now with CMOS you could; that's the difference. CMOS allows you to individually access each pixel on the chip. But even so, using each pixel in and of itself would result in black and white imagery. So you use pixels from all three chips, and you either use a spatial offset system (which makes the pixel calculations a bit more complex) or you end up with fixed-pattern noise in the video (if all three chips are perfectly aligned then you end up with "dead zones" between pixels on the sensors; spatial offset covers those gaps).

In short, it just ain't that easy.
 
Last edited:
Wow man. BArry where did you learn that? are you an engineer? did you have a mom? or were you assembled somewhere?

I just like cameras. And i really like the HVX. And i would love ot have one. 960X540 chips and all. cause the footage looks amazing. Simply put.
 
Dude you are so getting quoted! LOL

WEll just think barry, your encouragement of a young videophile 3 years ago led me to this point! :beer:
 
My that was obscenely detailed. Thanks Barry.

Your no-camera thing was interesting since I remember Adam claiming 1200 lines of horizontal resolution for the Cine Alta. Ah well, now that I have a clear answer i can go back to pretending resolution charts don't exist and just shooting pretty pictures.
 
DavidBeier said:
Ah well, now that I have a clear answer i can go back to pretending resolution charts don't exist and just shooting pretty pictures.
Click to expand...


AMen to that.
 
DavidBeier said:
Lets start with the HVX200. It's 960x540 chips pixel shifted to 1440x810, uprezzed to 1920x1080 and then subsampled at 1280x1080.
Click to expand...

Your math appears off... but as I understood it, the green CCD of the HVX was offset by 0.5 in both vertical and horizontal. 960 x 2 = 1920. 540 x 2 = 1080.

where did you you get 1440x810 from?

the image is in fact a true (ish) '1920x1080' image, albeit it loses true 4:2:2 colour this way..

am I wrong here? Have I been misled...
 
DavidBeier said:
What's holding you back from gettin and HVX Tim?
Click to expand...


Money mostly. im very poor. not good enough a DP to get real pay yet.
 
Last edited:
Loki said:
Your math appears off... but as I understood it, the green CCD of the HVX was offset by 0.5 in both vertical and horizontal. 960 x 2 = 1920. 540 x 2 = 1080.

where did you you get 1440x810 from?

the image is in fact a true (ish) '1920x1080' image, albeit it loses true 4:2:2 colour this way..

am I wrong here? Have I been misled...
Click to expand...

not that i necessarily understand any of it, but I've heard the 1440x810 number from Barry a FEW times. And Barry's word is law :)

I believe the 810 is AFTER pixel shift, and it's upscaled through software from there.
 
Loki said:
Your math appears off... but as I understood it, the green CCD of the HVX was offset by 0.5 in both vertical and horizontal. 960 x 2 = 1920. 540 x 2 = 1080.

where did you you get 1440x810 from?

the image is in fact a true (ish) '1920x1080' image, albeit it loses true 4:2:2 colour this way..

am I wrong here? Have I been misled...
Click to expand...
Pixel shift only gets you 1.4x or 1.5x the resolution, not 2x.
 
But you actually notice the difference between the two?

It cant be THAT big of a difference if its a little downrezzed...
 
Loki said:
Your math appears off... but as I understood it, the green CCD of the HVX was offset by 0.5 in both vertical and horizontal. 960 x 2 = 1920. 540 x 2 = 1080.

where did you you get 1440x810 from?

the image is in fact a true (ish) '1920x1080' image, albeit it loses true 4:2:2 colour this way..

am I wrong here? Have I been misled...
Click to expand...

At it's best, pixel shifting can give you as much as 1.5 times more resolution which is where 1440x810 comes from. That number is then uprezzed to 1920x1080 where all image processing is done (color, noise reduction, ect). At that point, the image is downrezzed and subsampled in 1280x1080 for 1080i, 960x720 for 720p and good old 720x480 in 480i.
 
The CCDs are offset by 1/2 pixel, giving 1920x1080 photosites to sample from. However, as others have pointed out, the best you're going to get in true resolution increase is about 1.5x. So as far as discernible DETAIL goes, Panasonic estimates it to be about 1400x810 (or thereabouts; there's a technical paper on their website that lists the exact numbers).

So you have about 1400x810 worth of detail being carried within a 1920x1080 wrapper. All internal processing is done using the full 1920x1080x60p matrix. That 1920x1080 then gets scaled to 1280x1080 for recording in the US version, or 1440x1080 in the EU version. Basically the same process that the new Sony is bragging about.
 
This is also really confusing to me so your saying an hd camera that supposed to shot 1080x1920 really only shoots shots 670x730....what?.....sd is 720x480 though so how does this all work....if i tested my dvx for resolution what kind of numbers would that give me?...and why does it still look sooo much better?...im kinda confused....haha sorry if my questions are stupid this is an issue I have never really even thought about.
 
The numbers do not correlate. They're not the same. The # of pixels in the frame has only a vague relationship with the # of lines of resolution the camera can resolve.

A standard-def camera like the DVX will clock in at somewhere like 400x400; or 400x550-ish in progressive/THIN line detail.
 
You must log in or register to reply here.
Back
Top