(pics) 16:9 Compression Tests - interesting results!

[Rob Hester]

its 5:42. i must be mad...
Hypothesis:
Using in-camera blacking bars makes less data to compress to DV, therefore less artifacts.
Rob's conservation of..er..DV if you will

Test Materials:

• 1 JVC DV3000U
• 1 printed resolution chart on dorm corkboard
• 1 mini tripod

The Test:
Shoot a resolution chart in 4 different modes and crop to 16:9:[list=A][*]cinema (black bars added) -interlaced[*]squeeze (meant for 16:9 projection) -interlaced[*]d.wide (meant for .7x built in wide angle - marketing lie - only gets ".7x" by using more of the ccd) -progressive[*]d.wide w/personally created title (my own black bars added) -progressive[/list=A]Results:

• Both d.wide shots show a greater capable resolution
• The rob-made title shows less compression artifacts (not much less...but still less)

[list=A][*]4:3 Cinema Mode
http://static.zed.cbc.ca/users/r/rhe...es/icinema.png[*]16:9 Squeeze Mode
http://static.zed.cbc.ca/users/r/rhe...s/isqueeze.png[*]4:3 Full D.wide Mode
http://static.zed.cbc.ca/users/r/rhe...iles/pfull.png[*]4:3 D.wide Title Function - created-own-black-bars-to-be-applied-in-camera-Mode
http://static.zed.cbc.ca/users/r/rhe...es/pcinema.png[/list=A]
Notes:

• Yes, i am trying to get the best out of cheap consumer equipment..and the differences may not be large - but they do appear to be there
• D.wide mode shots were zoomed back to 1x for equal distance - but the camera never moved
• Shots posted are in PNG file format

Conclusion:
The hypothesis proved true for me, and I share my information in hopes that others may find it valuable...i really hope someone comments or i'm gonna feel mighty stupid

As always, thanks for your time!!
Rob

UPDATE: Please check out the zoomed shot in my lower post.

[Rob Lohman]

It might be true, but I'm wondering about the DV compression
algorithm. I do know that it is using a fixed datarate and I do
believe it using a set number of bits per 8x8 datablock. So it
shouldn't yield much of a difference if any at all.

Now if you are shooting electronic 16:9 then your pixels get
stretched before getting compressed and thus you get a higher
bandwidth per pixel which should reduce compression "issues".

Not sure whether my first paragraph is correct. Been a while
since I looked at the technical side of DV compression.

[Glenn Chan]

It looks to me like d has less artifacts than c on the diagonal line.
d and c don't look at sharp as a and b in the corners.
There's some weird double images in all of the pictures. There's extra lines (kind of faint) where there shouldn't be any.

Adam Wilt's DV FAQ says that Canons get more resolution when you shoot 16:9 in camera. http://www.adamwilt.com/DV-FAQ-etc.html#widescreen

[Rob Hester]

Originally posted by Rob Lohman : It might be true, but I'm wondering about the DV compression
algorithm. I do know that it is using a fixed datarate and I do
believe it using a set number of bits per 8x8 datablock. So it
shouldn't yield much of a difference if any at all.

I could be entirely wrong, it may end up being the case :(. I have little knowledge of the algorithm, i took a guess that it would be close to JPEG - the colour information you have to process, the smaller the size, therefore you can conserve the fixed data rate for the part of the image which is valuable to you.
-----------------------------------------
UPDATE: Enlarged 400x comparison
http://static.zed.cbc.ca/users/r/rhe...les/zoomed.png
-----------------------------------------
So, with this zoomed in capture I have posted now, I believe that when you say "if you are shooting electronic 16:9 then your pixels get stretched before getting compressed and thus you get a higher bandwidth per pixel which should reduce pression "issues". you are correct.[list=1][*] If you look at the differences between only the two interlaced captures, the "16:9 Squeeze Mode" exhibits slightly less compression artifacts[*] However, this also plays out for the two progressive shots which are ultimately using the same # of pixels in favour of the "4:3 D.wide Mode with Title"[/list=1]

Originally posted by Glenn Chan: d and c don't look at sharp as a and b in the corners. There's some weird double images in all of the pictures. There's extra lines (kind of faint) where there shouldn't be any.

[list=1][*]I noticed the ghosting too - not sure what to attribute that to...poor glass? poor lighting conditions?[*]I don't know if d and c are less sharp though, look at the new zoomed image I made for you. This is a zoomed portion of the resolution chart (top right). If you look at where the lines converge, both D and C hold more true closer in than A or B. Well, in my opinion I guess. [*]I know that some cameras have better squeeze modes - my JVC just chops off the top and bottom of the picture as far as I can tell.[/list=1]
Anyways, thanks so much for the replies guys! hopfully there is enough intrest that maybe someone would want to try this on another brand camera?
ps. i know the format of this post is a little anal..but hopefully doing so makes things clear and concise?
Rob

[Glenn Chan]

You know that the Kell factor affects your results right? (if you're just reading the numbers on the side corresponding to where there are distinct lines) If the lines land in between pixels then they aren't really going to show up as one distinct line. That's basically what the Kell factor is.

I'm not sure if C and D are actually sharper. They have less contrast which makes the edges appear less sharp.

[Rob Hester]

You know that the Kell factor affects your results right?

Sorry, I don't know what you mean...could you explain it a bit more? I was looking at the distinct lines from the edge of the circle inward until the point at which they converge and the details are not enough so that you can determine whether or not the lines are stuck together or independant. I wasn't reading the numbers.

I'm not sure if C and D are actually sharper. They have less contrast which makes the edges appear less sharp.

I assure you, if I were to alter the contrast it wouldn't give me more pixel depth.
Also, look at the artifacts around the circles, which are more apparent? I cannot attribute all of that to contrast

I don't want this to turn ugly...not meaning to sound that way if i am Glenn! I'd just like to explain the results I found in the most scientific way possible.

[Rob Lohman]

What I meant was, that adding black bars shouldn't give you
more bandwidth for the remaining pixels. Using an electronic
16:9 stretch should, though.

[Rob Hester]

thats what I guessed you meant. I can understand if you are spreading the data out across 720x380 pixels as opposed to 720x480 there would be less compression needed. Edit: What was I thinking...squeeze still uses the full vertical res, just a different pixel aspect ratio 1.2 vs ..9, right?

The whole idea of this was to figure out if solid black takes up as much information as regular, multicoloured picture...I don't think it does by the looks of things...but I am definately inclined to believe that you are right.

What do you think about the captures I took though? If you only had to judge from those?

[Glenn Chan]

Sorry, I don't know what you mean...could you explain it a bit more? I was looking at the distinct lines from the edge of the circle inward until the point at which they converge and the details are not enough so that you can determine whether or not the lines are stuck together or independant. I wasn't reading the numbers.

Here's one site explaining the Kell factor: http://members.aol.com/ajaynejr/kell.htm
If you do the test again, I think you'll get different results because of the Kell factor.

I don't want this to turn ugly...not meaning to sound that way if i am Glenn!

I'm not trying to make it turn ugly either. :)

[Ted Springer]

It really depends on the individual camera.

Anyway, here's another theory for you, basically just what the other guy above me said (this version of vBulletin is out of date so I can't see the name of the person who posted it as I type this): Shooting in letterbox mode gives LESS data to the image since in "squeeze mode" it takes up the full 720x480, thus more "bits and bytes" for the same amount of image.

HA!

:)

[Rob Hester]

Ted, are you referring to Glenn? I don't think he was talking about squeeze mode, just the idea that the real determination of resolution needs to take into account the "kell factor" - which I partially understand.

to Glenn: I am curious now if you are right...I'm gonna whip up a new batch for ya, i'm interested in the final outcome.

Thanks for sticking around guys!
Rob

[Rob Hester]

UPDATE:Okay, here are the updated (or re-try) results...this time I added a few in, just for comparison

Notes:

• If you remember, D.Wide has .7x, I tried my absolute best to get the zoom correct...but there is a slight difference - the interlaced images are slightly closer
• The interlace and d.wide shots have different brightness levels - this is just what auto chose on both...if you iris lock in one, the other looks MUCH different - auto balances them out, but not quite perfectly as you can tell
• Because of this, I added an edited shot, the d.wide w/title corrected in brightness and contrast to at least match the look of the interlaced shots
• In the zoom comparison, I added two edited shots: D.wide+Title with Contrast, and D.wide+Title and the Unsharp filter

[list=A][*]4:3 Cinema Mode
http://static.zed.cbc.ca/users/r/rhe...s/3icinema.png[*]16:9 Squeeze Mode
http://static.zed.cbc.ca/users/r/rhe.../3isqueeze.png[*]4:3 Full D.wide Mode
http://static.zed.cbc.ca/users/r/rhe...les/3pfull.png[*]4:3 D.wide Title Function - created-own-black-bars-to-be-applied-in-camera-Mode
http://static.zed.cbc.ca/users/r/rhe...s/3pcinema.png[*]4:3 D.wide Title Function + Contrast & Brightness Adjustments
http://static.zed.cbc.ca/users/r/rhe...macontrast.PNG[*]400x Zoom Comparison
http://static.zed.cbc.ca/users/r/rhe...iles/zoom2.png[/list=A]

This sorta did take a while...comments are GREATLY appreciated!
Thanks guys,
Rob

[Rob Lohman]

Yes the resolution will still be 720x480, but, each original
pixel (BEFORE the stretch) will be larger after the stretch (some
pixels will be dropped because they fall outside the 16:9 area)
and thus have more bandwidth available per pixel.

[Rob Hester]

Yeah that makes perfect sense...but why then does it still appear to have more artifact? (look around the numbers and middle circle) Is this the cause of the squeeze mode having a greater sharpening effect applied to it maybe?

[Rob Lohman]

I'm afraid I don't have the answer. It will probably be different
per camera. What algorithms they are using and quality of the
algorithm / components etc. Hard to say.