real resolution of CF30.
 

Man am I going to get my but kicked for this one...

I was playing around with raw still images from the Z1 the other day. I was trying different ways of processing the footage to see what makes it tick.

Now please do not take this as an attack on the Z1. I think it is a great camera and it is very high quality when compared to SD and even other forms of HD.

That being said I decided to do one test where I compared a CF30 still frame to that same exact frame scaled down to 960x540 and then back up to 1440x1080 again. The results were almost identical to the raw 1440x1080 still image. There was no way at all to tell between the two images. I put both images on top of each other in Photoshop and clicked the top layer off and on to try and see any changes. After about an hour and only with certain images the only change was about a 0.1 pixel blur and that was only in certain areas of the images. What I mean by a 0.1 pixel blur is if you were to apply a gaussian blur in Photoshop with a value of only 0.1 pixels instead of 1 pixel. The only real way to tell any change at all was to use a difference transfer mode and see if any pixels were left behind. There were a few but the results were very very small.

Try it out.

What does this mean? Well to me it means that even with using pixel shift at the end with CF30 or CF25 you only really end up with the detail of a 960 x 540 pixel image. Pixle shift only gives you a slight boost in sharpness but nothing you would ever notice on a monitor. Better de-interlacing methods might help as well.

Does this really mean anything? Not really. I just thought it was interesting. I still think the Z1 is a great camera. I actually think slightly better of it knowing what it can do with that CCD block. However you look at it that 960x1080 CCD block is a very good one.

It does however bring up an interesting point as to how a HDV camera with a native 720p CCD block at 1280x720 could look in terms of detail compared to a Z1. Of course this only really matters if you use a CF mode. If that 1280x720 CCD block isn't well made then the 960x540 might look just as good if not better.

Again this is not an attack on the FX1/Z1 but just an observation. I actually have more respect for the camera seeing how good of a CCD block is actually on the camera. I would rather have a high quality 960x540 scaled up instead of a crappy 1280x720 block. If you start with crap you will always have crap. If you can start with a high quality image even if it is slightly lower resolution you will get better results with any changes you make.

If you want to see the changes in a more meaningful way, try to use the "difference" mode (in AE - not sure what it's called in PS) to subtract the two images. Any changes - particularly blurring of the edges shows up exceptionally well this way.

I'm not at all surprised by the 540 vertical resolution. I have long subscribed to the theory that CF30 simply takes a single field of exactly 540 lines and doubles it. I'm a little skeptical that you didn't see excess blurring in the horizontal dimension though, as I'm pretty sure pixel shift does something beyond 960... but it really depends on what was imaged.

I personally love CF30 for the workflow-related benefits.

-Steve

I tried your test, with a resolution chart shot in CF30. I can't duplicate your results.

While in CF30 it appears to be true that you're only getting one field of information, the idea that you can shrink the 1440 image down to 960 and then back up to 1440, losslessly, makes no sense to me -- and in my test of a resolution chart, there was significant loss of resolution in the reconstituted 1440, as opposed to the original.

Pixel Shift isn't some sort of post-processing up-rez technology -- it's genuine resolution captured at the time the image is taken. I wouldn't expect to be able to up-rez in post and match the actual resolution captured at the time the shot was taken.

But, maybe I didn't do the resize in the same way you did. Here's a .JPG of the original frame so you can try it yourself and see what results you get.

http://www.icexpo.com/FX1-CamAlign-CF30.jpg

As far as I have heard there are 3 960x1080 chips inside. That might be the reason why he scaled down to this resolution.

But this does not make too much sense, because the 1440 are composed by pixel shift. So the left/middle/right chip are sub/pixels off the others.

This is not true 1440 but it's better than 960...

I guess that Thomas didn't use technical charts. So it's not that obvious at the first sight. But the testpicture from Barry is proofing the oposite. I tried it mysellf.

But I really have to say, that scaling a SD image up (no test charts) isn't that far away from HD when viewed without direct comparison.

IN comparison I am still blown away from the difference of HD.

But we had a premiere of our last shortfilm in the ARRI headquarter cinema, which means to me it's the absoulte TOP beamer, and the SD signal from that movie looked astounding on the big screen.

Still I am a HDV believer and know we are making sharper pictures that way :)

Peter

Thanks for the image Barry. I so want this to be wrong.

Well I did the same test and I got pretty much the same results.

1. take the image and duplicate in photoshop.
2. Scale down the copy to 960 x 540.
3. Scale up the copy back to 1440 x 1080.
4. Select all and copy the image and then paste onto of the first image.
5. Look at the two layers by turning off the top layer or using a difference transfer/blend mode.

There is a very slight change but only about the same as if you were to add a 0.1 pixel blur to the image. I think this is the only real amount of extra data you get from pixel shift. If pixel shift actually worked than the Canon XL1 would look as sharp as a SONY PD-170. The fact is that the XL1 would never have the same level of detail as a CCD block with a high pixel count.

The pixel shift on the FX1/Z1 does seem to give a tiny bit of extra detail but not 1440 pixels worth of it. Then again I am using a bi-cubic filter twice so that could be the source of the very slight blur which would almost mean pixel shift isn't doing much of anything.

When DV camcorders first arrived they often used left-over Hi8 CCDs of about 270K when at least 380K was needed. I call this under-sampling. In short, I say a camera undersamples when the CCD offers less native rez than the recording format. So the Z1, IMHO, is an undersampled camera -- which is why I see little difference between it an a 1280x720 progressive camera.

I think when I (we) claimed green-shift yielded a 1.5X increase in luma rez -- we were guessing. Geee-- 960x1.5 = 1440. I think the increment may be far less. Maybe only 1.25X. Which is why, despite being beat about my ears -- I've maintained the FX1/Z1 simply is not very detailed.

CONVERSLY, one person at Sony claims pixel-shift yields 1920. Moreover, they claim the DSP chip handles 1920. Turns-out they are right of the DSP. Looks like it handles 1920x1440. In fact the CMOS camera uses the same DSP and will record REAL 1440x1080. Of course, that doesn't mean the lens will handle the extra rez, but we could might get "better" HD from a $2000 camcorder.

SEE: http://www.sony.jp/products/Consumer...HC1/index.html for lots of detail. This is one cool cheap HDV camcorder!

Of course, after Sony's detailed briefing on the CMOS+IP chip -- it is a crying shame 24p wasn't added. So if JVC can create the same kind of camera, but offer progressive -- there will be another low cost HDV camcorder later this year (IBC?).

The best reference to pixelshift I've found either quote a theoretical max of 1.414 or the BBC who say about a 30% increase in resolution is about right. Either way, I think lens limitations are getting in the way of seeing either of these figures.

Thomas's experiment is interesting, and yes, it doesn't stand up too well to a test image, but on any naturalistic image, it really doesn't look much different at all. It's not night and day, but subtle.

Graeme