Wide Shot Resolution

[Boyd Ostroff]

<<<-- artefacts like blocking, quilting, mousquito noise... are the DV shortcommings. -->>>

And that's what I always had heard as well. But in my test these do not appear in the JPEG of the chart that was compressed with the DV codec however the do appear in footage of the chart that was filmed with my VX-2000.

Whatever the reason, these artifacts must be related to the camera's chips, hardware or firmware. But to put this all in perspective (trying hard to snap out of my depression here ;-), that was a tiny area of the full image which was enlarged 300%. The full chart from the VX-2000 looks like this. But what surprised me was how just how clean the DV-compressed JPEG looked. To me, the main difference between it and the original is the limited luma range which shows up in the darker gray background.

[Dan Uneken]

Just watching Eurochampionship football on TV.
Wideshots of the entire pitch: the players show up like blotches, definitely soft.
Close shots are brilliant in definition.
Same thing. Big expensive cameras. (New TV)

[Andre De Clercq]

Boyd tell us somewhat more about those enlarged pics. I suppose the first one is an original (electronic) JPEG image of the chart. Size ? (KB). The second one, recompressed in the DV format? Full frame? The thirth one: a VX2K pic taken from the chart?.Looks to me as a line doubled field (low vertical resolution).

[Richard Maloney]

Interesting. I think that Rob is on to something, makes me wonder if (non-mechanical) image stabilization would be part of the cause of this degradation?

[Boyd Ostroff]

Andre: For the camera test I used the chart from John Beale's website which is 3458x2608 pixels to print 216 dpi hardcopy at 16"x12" on high quality inkjet paper with an Epson Stylus 1280. Shutter speed 1/60 sec, manual iris with zebra to determine exposure. Camera was ~10' from the chart. I shot that over a year ago however, I wish I was viewing the full frame at the time because it could have zoomed in slightly. I think the VX-2000 image is in the same ballpark as tests I've seen elsewhere:

http://www.pixelmonger.com/chrt_pd150.jpg
http://www.bealecorner.com/trv900/respat/trv-res.jpg

My JPEG on the left is the same file used to print the hardcopy, but resized in photoshop to 720x480. The middle one is that JPEG dropped into an FCP sequence, exported as a Quicktime DV-compressed file. I then opened the movie in the Quicktime Player, copied a frame and pasted into photoshop.

All of the images are 720x480 and I just enlarged and cropped a representative section to put online.

[Rob Lohman]

Richard: if it is a non-optical image stabilizer, then yes, it will
definitely affect quality and/or resolution.

[Andre De Clercq]

Thanks Boyd for the additional info. Still two questions:
-Did you drop the VX2K picture (image sequence I suppose) also into FCP and further followed the same procedure as for the middle pic?
-Is QT outputting (deinterlaced) frames?
I am still puzzled about yr results: allmost no dif between a 3458x2608 JPEG and a 720x 480 DV image for the left two pics, and the large amount of mousquito noise and quilting on the right pic.

[Boyd Ostroff]

Andre: as I said, the VX-2000 clip is a little old, but I actually filmed to tape, captured in FCP then exported as a JPEG. Sorry, I don't remember further details. Maybe later today I'll capture another example from both my PDX-10 and VX-2000 under more controlled conditions and see if there's any difference. But like I said, try enlarging those charts in the other links I provided. I see the same sort of mosquito noise there as well.

And just to clarify, the hard copy of the chart was 3458x2608. The two JPEG images on the left were first resized to 720x480. But regardless of the quality in the VX-2000 example (as you note) the surprising thing is how clean the DV compressed version of the original JPEG is.

[Boyd Ostroff]

OK, I re-shot my test chart. I also tried the same test with my PDX-10 and included both results. The new test can be seen here.

This time I was careful to frame the shot as exactly as possible so the chart filled the whole frame. I was about 6' away and shot outside in the shade with indirect light on a cloudy day. The video was captured directly via firewire to my laptop running BTV Pro. I then dropped these clips in FCP and exported still images as JPEG's at the highest quality. I did not deinterlace.

I used manual white balance to a blank sheet of the same paper as the chart printout (which was again 16"x12" @ 216 dpi). I zoomed in to the max and manually focussed, and tried autofocus as well with no apparent difference. Then I zoomed back out to fill the frame with the chart. One interesting thing, the VX-2000 couldn't seem to white balance properly under these conditions, the result was very bluish. Both cameras were shooting interlaced at 1/60 sec shutter. The VX-2000 was around f6 using the ND1 filter. On the PDX-10 it doesn't tell you the exposure and it automatically drops in its undocumented ND filters. I converted the results to grayscale and used the Photoshop auto levels function. Once again I cropped the same area of the chart and enlarged 300% with Photoshop bicubic interpolation.

So anyway, these tests yielded better quality images and dont have the blocky artifacts of the old test (I'm guessing the problem with the old image was that JPEG quality was set low when I exported from FCP). However they don't really change the conclusion that the camera is the main limiting factor in image quality and not the DV codec.

[Andre De Clercq]

Thanks Boyd for the interesting pics. Cams do indeed limit resolution. But that second (non cam) pics still puzzles me...Did you really drop the resized JPEG testpattern into the FCP timeline, made a DV rendered AVI file (to get the DV conversion) , and exported a frame out of this AVI file?

[Boyd Ostroff]

Yep, although it was Quicktime... isn't AVI some sort of PC thing? In fact, I tried the test two different ways since the first time it looked too good to be true. I was afraid that FCP was using some higher-quality internal representation rather than DV when I exported directly from the timeline. That's why I rendered and exported as a DV-compressed Quicktime movie, then used the Quicktime player (a separate application) to copy the frame and paste into Photoshop. I got similar results using this same methodology with a photograph of a landscape. And that's what got me started on this whole thing. I was pretty sure that if I took a clean JPEG of a landscape with lots of fine detail that it would fall apart under DV compression. Much to my surprise it did not. Guess I need to start saving for a DSR-570 ;-)

Try some tests for yourself and see if you get different results...

[Ignacio Rodriguez]

Shocking!
Where is all that noise coming from?
Where is all that resolution going?
Boyd, is it possible that the 'sharpness' setting on your VX2000 and PDX10 are set way down? Could it be that we are setting them too far down? Notice: I include myself. Also, is it possible that the AE point is also set too low? I am sure you probably set it down too, like I do, so the image might be somewhat underxposed to begin with, this might account for the noise. Also, do you have FCP set up do it's math on the maximum bits per channel?

[Andre De Clercq]

Boyd I did some tests the in the past using the eia1956 and zoneplate testpatterns, maybe I'll have to do this again, but I never got a pic out of a DV stream that good. What is the filesize for yr DV converted JPEG pic? As far as I remember it should be less than 80KB.

[Boyd Ostroff]

Iganacio: I always keep my PDX-10 sharpness at the minium, not sure on the VX-2000 and it isn't handy right now, but I suspect it was a notch above the minimum. You are probably right about this influencing the results, hadn't thought of that. However most of the time my goal isn't to produce sharp resolution charts, and after lots of experimentation I like the image much better without the artificial edge enhancement, which also accentuates any noise. I don't think there is any setting for computational precision in FCP. And regardless, FCP was also used for the DV compressed JPEG so the playing field would be level.

Andre, I'm at work so I don't have those images handy to check file size. By all means, don't take my word for any of this though! Run your own tests. This is the chart I used. I'd be interested to see if you got different results using Premiere, Vegas or another NLE.

[Ignacio Rodriguez]

Oh there is. I am not near my Mac right now, but I am sure you can change the bits per channel somewhere in FCP4. The point is, this might make a difference, quantization noise increases noise, ar at least apparent noise, when you start out with a noisy image. Also I think underexposing in the camera might also have an effect on noise. I would love to do the tests myself, but I don't have a printer, don't have a VX2k, don't have a digicam, except the PDX10 itself... which gives me another idea: you could include in the test the PDX10 in photo mode... and thus see how different the CCD array operates in that mode, or not.