Why shoot 1080p on HVX200?

Status
Not open for further replies.
So why not just shoot 1080 then and skip the time consuming software uprez. Seems to make the most sense to shoot the higher res in order to get the higher res. You'll never get more resolution out of 720 than the 1080 footage. Getting something out of nothing is gonna be kinda hard... And considering the hvx doesnt upres the 1080 footage but instead shoots 1080 in the signal chain no matter what and down res's per the format I would think shooting 720 and uprezzing would be quite the waste of time. Might want to do some research into how the camera works.
 
bill totolo said:
I believe the native resolution of the camera is 960x540. The upscaling to 1080 is pretty gnarly and (I think) better executed in post production. I just shot a commercial that will air in theaters and tested 720 v 1080 blown up to 42" using test charts.

Also just for fun I compared this to the Sony F900. Admittedly it's not a fair test and, as expected, the results weren't close.

Calling the HVX a 1080 camera is slightly misleading, I'd say. IMHO.
Click to expand...

Bill you are un-informed as to how the images are made, there is no upscaling.. The signal off the CCD is captured at 1080P. The images presented to the capture have the equivalent of a 1.1 million pixel imager, this is after the spatial offset gains. All of the camera processing is done in the 1080P damain and then the cross or down conversion happens just before recording.

Best,


Jan
 
Jeff Anderson said:
So why not just shoot 1080 then and skip the time consuming software uprez.
Click to expand...
Why NOT to shoot 1080 on an HVX200:
1. Reduced recording time, with no benefit
2. Higher image compression (lower image quality) due to larger frame size
3. Need to reverse-pulldown the footage to recover 24p

And considering the hvx doesnt upres the 1080 footage but instead shoots 1080 in the signal chain no matter what and down res's
Click to expand...
It is impossible to produce a 1080 line picture from three 540 line CCDs without interpolation, even WITH pixel shift.

Let me say it another way:

THE CAMERA IS INTERPOLATING THE DATA FROM THE CCDS TO GET 1080.

Definition of the word "interpolation:"

"In the mathematical subfield of numerical analysis, interpolation is a method of constructing new data points within the range of a discrete set of known data points."

In the case of the HVX, the discrete data points are the positions of the red and blue CCD photosites, and the offset of the green photosites. This data is not useful without interpolating the spatial difference between the CCDs.

The F900 is an example of a camera that DOESN'T interpolate. Why not? Because the CCDs are ALREADY 1920x1080. And not surprisingly, the resolution of the images from the camera are superior to the HVX.

I agree that there is a benefit to pixel shift, as it helps the upscaling algorithm make better decisions as to what the interpolated information should be, but the fact remains that you cannot truly resolve 1080 lines with this CCD setup. THERE ARE NOT ENOUGH PHOTOSITES.

Might want to do some research into how the camera works.
Click to expand...
Seems as though I'm the only one that has.
 
Dude seriously Chill out.

All i know is th 1080P looks ****ing Good. Period. Go buy a EX1 if youre unhappy.
 
What's going on?
Nobody knows? All these threads are horribly confusing.
Jan, help us.
Tell us the truth.
 
Jan_Crittenden said:
The signal off the CCD is captured at 1080P.
Click to expand...
Clever wording, but this doesn't really tell us anything. "Captured" could mean anything. In this case, I assume it is what is being fed to the image processor in the camera, which will either 1. scale 1080 lines to 720p, 2. Apply 3:2 pulldown, or 3. interlace 1080 lines to 59.94i.

However, the data coming off the A/D is raw sensor data, which must first be resampled to 1080. Each CCD will produce one 960x540 monochrome image. It is the responsibility of the image processor to assemble these, using a resampling algorithm (similar to a demosaic algorithm for Bayer CMOS data), into a 1080p image. This obviously cannot be done without resampling, as the images are 960x540! Even if you have three CCDs with overlapping photosites (pixel shift), you still must make assumptions about the location of image elements, of which parts exist on one or more offset pixels!

The F900 avoids all this because its CCDs already produce, natively, a 1920x1080 monochrome image. There is no pixel shift, no resampling.

I will use Andromeda / Hydra as my example: these systems grab the RAW CCD DATA FROM THE A/D and bypass Panasonic's image processor, applying their own resampling / scaling / interpolating (and utilizing pixel shift). Again, this is how the DVX100 could get 720p! IS THE DVX100 A 720P CAMERA? NO!
 
Thomas Worth said:
The F900 avoids all this because its CCDs already produce, natively, a 1920x1080 monochrome image. There is no pixel shift, no resampling.
Click to expand...

The F900 is also a $100 000 camera, I think its safe to say that we've seen what the HVX is capable of, and @ $5000 its pretty spectacular.
 
Last edited:
Thomas Worth said:
It is impossible to produce a 1080 line picture from three 540 line CCDs without interpolation, even WITH pixel shift.
Click to expand...

You are just simply incorrect. If you had an understanding about cameras, signal processing and phsics, you might understand, but you seem to think that you are right without a foundation for the technology that drives this. I have engineers that might argue this with you.

Let me say it another way:

THE CAMERA IS INTERPOLATING THE DATA FROM THE CCDS TO GET 1080.
Click to expand...


And again you are incorrect. There is no interpolation. The picture presented to the DSP is taken at 1080P.


Definition of the word "interpolation:"
"In the mathematical subfield of numerical analysis, interpolation is a method of constructing new data points within the range of a discrete set of known data points."
Click to expand...

But this would apply if that was what we were doing. It isn't.



The F900 is an example of a camera that DOESN'T interpolate. Why not? Because the CCDs are ALREADY 1920x1080. And not surprisingly, the resolution of the images from the camera are superior to the HVX.
Click to expand...

And not suprisinly the cost of the F900 is about $96,000 more.

I agree that there is a benefit to pixel shift, as it helps the upscaling algorithm make better decisions as to what the interpolated information should be, but the fact remains that you cannot truly resolve 1080 lines with this CCD setup. THERE ARE NOT ENOUGH PHOTOSITES.
Click to expand...

You need a better undersatanding of how the CCD works, a CMOS works they way you are thinking, but CCDs have an analog output.
 
Thank you Jan.
So what do YOU think: What's better 1080 24p or 720 24pn?

If I want to transfer to 35mm film, what would be my choice?
 
OK, fine, I give in - I had resisted getting involved in this particular little spat until I saw this:

You need a better undersatanding of how the CCD works, a CMOS works they way you are thinking, but CCDs have an analog output.
Click to expand...

...which is quite the worst piece of drivel I've ever heard out of Jan, which makes this something of a special occasion. The presumptiveness inherent in questioning someone of Mr. Worth's expertise with dross at the level of a Panasonic internal Powerpoint briefing is just mindblowing.

For the record, the joys and challenges of this sort of interpolation are precisely the same between a CMOS and CCD imager of otherwise equivalent specification - CMOS is a semiconductor manufacturing process which allows for certain conveniences, such as putting the output amps on the same substrate as the photosite array. The most common downside to this is that putting other circuitry around the photosites reduces fill factor, which reduces sensitivity and dynamic range and exacerbates aliasing, but the upside is that it means things like Red get to exist.

It does not, however, alter the challenges of interpolation one iota.

You can call what an HVX does interpolation or resampling or whatever you like; it's a process of inferring likely data points which don't exist based on ones that do. The question here seems to be one of how well that can be done given various forms of input data - three CCDs, a Bayer array, three offset CCDs, whatever. When you're doing Bayer demosaicing, for instance, you can start to heterodyne additional information out of the various frequency components of any axis of the photosite array - often horizontal, vertical, and the two diagonals are done, for instance. Similar things apply to pixel shifting, although there's a question over how much of this is really being done in a camera which costs what it does, fits in your pocket and runs for hours on one battery. This is not to criticise the thing; this is what makes it affordable and useful for the things it's good for, but let's not assume it's doing a NASA grade mathematical job.

Characterising exactly how well this sort of thing really works is difficult, but we know a few things empirically. For a start, green channel pixel shift doesn't work at all on objects which contain no green; they will be imaged at the native resolution of the R/B channels. The reason it's done that way is because the camera is only recording a 4:2:2 signal anyway and the only full resolution component is seventy-plus-percent derived from the green channel, mainly because video green is not very saturated in the grand scheme of things and we see detail mainly in luminance.

This works fine until you start to try and do postproduction work on it; chromakeyers see in colour, not luminance like we do, so these assumptions start to break down and the idea that this is really equivalent to a 1080-line device starts to crumble.

There is nothing intrinsically wrong with any of this; it's part of the set of compromises that have been made to create the HVX with a given feature set at a given price point, but let's not pretend it isn't interpolation. I consider the HVX-200 to be a mediocre 720p camera - mediocre mainly because it's noisy, but that's a side-effect of making the chips go so fast, and I should point out that's in comparison to things like Varicam. For the price, though, it's a fine piece of kit from a sheer picture quality point of view, and does things you can't get elsewhere without spending huge quantities of money.

Why can't you guys just be happy with your 720 line, 60-frames-a-second cameras? Why do you have to try and support daft ideas about it?

P
 
Phil Rhodes said:
Why can't you guys just be happy with your 720 line, 60-frames-a-second cameras? Why do you have to try and support daft ideas about it?
Click to expand...

Ummm . . . leaving aside the extremely tired rehash of the resolution fight, why do you make a point of "60-frames-a-second" here?
 
Phil Rhodes said:
The presumptiveness inherent in questioning someone of Mr. Worth's expertise with dross at the level of a Panasonic internal Powerpoint briefing is just mindblowing.
Click to expand...
Thanks for saving me the trouble, Phil. :thumbup:

I do want to add this interesting little tidbit, pulled from Wikipedia (http://en.wikipedia.org/wiki/HVX200):

The camera uses three 1/3 in 960x540 pixel (0.5 mega-pixel) progressive scan CCD's to capture the image. HD resolution is achieved by both horizontal and vertical spatial offset (aka. pixel shift or 'SBS-HD'). The green CCD in the array is physically shifted 1/2 pixel biaxially to achieve up to 50% higher horizontal and vertical resolution. This would make the theoretical maximum resolution of the image, 1440x810 pixels, even though each CCD has only 960x540 photosites. Internally the camera uses 1920x1080. In real-world use the actual measurable resolution of this camera has been tested at 540 lines horizontally by 540 lines vertically (see http://www.adamwilt.com/HD/4cams-part2.html).

The Adam Wilt link was especially interesting.

So Jan, let me just get things straight here. Am I to assume that Wikipedia, Adam Wilt, Phil and I are all completely wrong about this? I don't mind if you say we are, but I would appreciate some facts to back that up if so. You also mentioned having engineers that would debate the issue. Let's have one of them post here, as that may help to clear up this issue. That would be great! Thank you so much for offering!
 
Thomas Worth said:
Am I to assume that Wikipedia, Adam Wilt, Phil and I are all completely wrong about this?
Click to expand...

The idea of taking a Wikipedia article as the definitive answer on ANYTHING (not to mention as an argument against Jan's knowledge) is laughable to the extreme. Time for some courses in scholarly research.
 
Hi,

And this should be the final post on the subject as there are more agendas than just the point of the thread which was, is there more res to have in 1080 than in 720. This happens to be true, that there is more resolution to be had. This is due to the 1080 part of the codec. Point is if you have the camera, a multiburst chart and a waveform monitor, you can do this test and see for yourselves. You can argue 6 ways to Sunday and it doesn't chage the outcome. There is more resolution to be had on the 1080P function of the HVX200.

You can post as many references as you like, as many anecdotal insights, whatever, the facts are the facts.

Best regards,

Jan
 
Status
Not open for further replies.
Back
Top