SOT: Digital Zoom

amendegw wrote on 3/3/2012, 3:17 PM

I saw a post in another forum suggesting that adding gain in the camera was a better idea than fixing in post. That made me think about the follow-on question. What about digital zoom? I've almost never used camera based digital zoom - figuring there was no advantage to it. I can always pan/crop (i.e. zoom) in Vegas and there may be some detail outside the zoomed area that I may want to retain.

Other thoughts?

...Jerry

Comments

farss wrote on 3/3/2012, 3:30 PM

Depends.
On some cameras digital zoom works before scaling so you can get better results using it in camera compared doing it in post. As always test, test, test.

I would just print out a resolution chart and compare the results with different amounts of digital zoom. The other thing is most video camera's lens go long enough as it is, once you get beyond that point it's difficult to get a good image from thermal shimmer in the air and dust and pollution.

Bob.

Laurence wrote on 3/4/2012, 12:58 AM

Also, digital zoom on some multi-megapixel DSLRs is sometimes refered to as "crop factor" and gives you a full1080p image because there are enough pixels in the center of the sensor so that you don't actually lose any video resolution. Examples of this are the Canon T3i and the new Nikon D4 and D800, you can digital zoom way in without losing any resolution.

richard-amirault wrote on 3/4/2012, 6:58 AM

The other thing is most video camera's lens go long enough as it is,....

Not in my experience. Most cameras don't go long enough. 10 or 12x is not enough for me. 20x optical is what I need (or more) and it limits the choice of cameras for me.

amendegw wrote on 3/4/2012, 8:01 AM

Okay, here are the results from the test Bob (farss) suggested. Footage was 1920x1080 60i AVCHD from my Panasonic AC130.

First, 22x in camera Optical Zoom:

Next, 22x in camera Optical Zoom + in camera 10x digital zoom.

Next 22x in camera Optical Zoom + Pan/Crop in Sony Vegas

Finally, 22x in camera Optical Zoom + Pan/Crip in Sony Vegas + yadif Deinterlace.

So, my conclusion is (at least for the Panasonic AC130). For interlaced projects with interlaced footage, the camera's digiital zoom is somewhat preferable to pan/crop in Sony Vegas. However, if one pan/crops progressive footage, there is very little difference.

...Jerry

johnmeyer wrote on 3/4/2012, 10:56 AM

Jerry,

Nicely done test! Thanks for taking the time to do it. I had always assumed (bad thing to do ...) that I'd get the same results, whether the camera did the extra zoom or whether Vegas did it, but when the OP raised the question, and Bob made his post, I realized that my assumption might be very incorrect. From what you posted, it looks like your camera does a great job with its digital interpolation, but that you can match it if you do some additional post-processing.

I guess it would still make sense to do the test for one's own camera(s), since each cam may have different capabilities. Unfortunately, I don't have a nifty test chart like yours, but I guess I could get a good approximation by using an appropriate real-world object.

amendegw wrote on 3/4/2012, 11:04 AM

"Unfortunately, I don't have a nifty test chart like yours..."Ha! You know what I did? I when here: http://www.belle-nuit.com/testchart.html to get the chart, loaded it on my USB Flash drive, went down to Costco and had them print it nominally 11x14 on nice stiff cardboard. It was several months ago, so I don't remember exactly how much it cost, but I'm pretty sure it was under $10.

...Jerry

Edit: After I posted this I had a scare. I re-downloaded the Belle Nuit Chart and zoomed in on the source .tif. Here's what I saw:

Was the aliasing seen in the diagonal in the 1 in the third picture above a better representation of the actual image?

Well, it turns out that there must have been anti-aliasing correction in Costco's printer software as here's picture of the area using the "macro" capability of my point-and-shoot from about 1" from the image:

A case of two "wrongs" making a "right" [grin]

btw: The above image was taken 1 inch from the chart whereas the camcorder zoom was taken from about 8 feet!!

farss wrote on 3/4/2012, 2:35 PM

Interesting results.
The ISO 12233 resolution chart is available for free from here
You may find it quite difficult to print it out with 2K resolution though. Printed copies are around at various price points.

Firstly I would not have used YADIF. As the camera was (hopefully) locked off hard then a simple field blend is all that's needed to de-interlace. I say this because exactly what YADIF is doing is unknown and it could be adding edge enhancement. It really is important to compare images correctly when doing these kinds of tests.

Even with YADIF though the difference is there and that's not that surprising when shooting interlaced. I'd suspect the camera does the digital zoom before it does line pair averaging. The loss of contrast is quite interesting.

What I would have done though is to move the chart further away so that with the digital zoom applied it was full frame. If you use the ISO chart the trumpets make it quite easy to see what is going on resolution wise and you can read actual resolution in lines.

Bob.

amendegw wrote on 3/4/2012, 3:00 PM

"What I would have done though is to move the chart further away so that with the digital zoom applied it was full frameHa! This was done in my basement "Studio" (boy, that's really a stretch! - it's a green screen and a couple of mounted lights). My original intent was to get a full frame, but my camera was about as far away as I could get it without moving bookcases, and junk - and then I might have doubled the distance to the chart (15-18 ft) - still not enough to get the full chart at full zoom. Oh, and the camera was tripod mounted for the shots.

...Jerry

farss wrote on 3/4/2012, 3:14 PM

"boy, that's really a stretch"

I kind of figured as much as I typed :)

I've seen Panavisions setup, the charts are mounted on a motorised gantry so at the push of a button they can be moved a known distance. Nice if you can find the work to pay the rent for the real estate involved.
All of this really is expensive to get into. A printed 4x chart from Edmund Optics is around $1,500 :(

Personally I can think of better things to spend my hard earned on, like a BIG silk for shooting outdoors and the crew to set it up.

Bob.

amendegw wrote on 3/4/2012, 3:43 PM

I've been thinking. I really don't need to zoom to full optical zoom to test this concept - just zoom to keep the full chart in frame, then click on the digital zoom.

Maybe a project for next week. We'll see. I've got some other stuff going on as well. Also, I may go to Costco to get a Hi-rez print of the ISO 12233 chart.

...Jerry

farss wrote on 3/4/2012, 4:15 PM

"Maybe a project for next week"

Just be careful my son. You could end up spending the best part of a decade as a measurabator like myself, not good, not a good way to go at all.
I shot my first narrative drama in 40 years a couple of weeks ago, what a challenge and what an eye opener. I doubt anyone else on the shoot had a clue what I meant when I said "That's a take, print it" but man it felt GOOD :)

Bob.

JasonATL wrote on 3/4/2012, 4:20 PM

Jerry - Interesting test and well executed.

One thing that occurs to me is that, in theory, in-camera zoom occurs prior to any compression. In Vegas, you are zooming in on the image and any compression noise.

One potential way to test the effect of this is to start with a static image in Vegas, say a 1920x1080 high quality jpg. Perhaps put some kind of key framing in it to cause motion, perhaps even change its color (I'm just trying to get the codec to have to do some work and not sure if this does it). Then, render it out with compression akin to what a camera would do. Now, take this same image and zoom in with Vegas (call this sample "A"). Compare this result to loading the previously-rendered regular sized file and zoom in with Vegas (call this sample "B"). If my theory is correct, "A" should look better than "B" and you've held constant the scaling capability (of Vegas). This at least helps determine if the scaling in-camera is due to a better in-camera algorithm or something else.