AVS Forum banner
Status
Not open for further replies.
1 - 20 of 109 Posts

· Registered
Joined
·
629 Posts
Discussion Starter · #1 ·
Hello everyone,


I did this little test to compare the different scaling errors introduced when deviating from the standard DVD resolution of 720x480.


What I did, was create a 720x480 grid of alternating black and white dots - similar to the test patterns found on VE and AVIA.

I then resized this grid in Photoshop using the "Image Size" control. Constrain proportions was turned off, and resampling was set to "bicubic".


Each resulting grid was magnified to 200% to show the scaling errors more easily. Here ya go:

http://members.socket.net/~ryandinan/720x480.jpg

Original 720x480 grid

http://members.socket.net/~ryandinan/1440x480.jpg

1440x480 (horizontal resolution doubled) - looks good - but spaces between the white dots have turned dark grey.

http://members.socket.net/~ryandinan/1440x960.jpg

1440x960 (doubled horizontally and vertically)

http://members.socket.net/~ryandinan/960x540.jpg

960x540 - Look at the mess it creates!

http://members.socket.net/~ryandinan/1920x1~1.jpg

1920x1080 doesn't look that bad, but tends to soften the edges of fine detail.

http://members.socket.net/~ryandinan/1024x768.jpg

1024x768 - Looks about as bad as 960x540


It appears that even multiples of the source resolution, provides less scaling errors, and gives a sharper, more detailed image.


Please let me know what you think - I'm sure the "test" has a problem somewhere - most likely in the resampling area - What method do most modern video cards use to resample the image?

If "nearest neighbor" was chosen, every image maitained sharp black and white dots - but they were just spaced unevenly - see below:

http://members.socket.net/~ryandinan/960x54~1.jpg

960x540 nearest neighbor resampling.


-Ryan Dinan
 

· Registered
Joined
·
6,227 Posts
Even multiples of the source make perfect sense for the CRT users who have the flexibility... I believe Ken has some nyquist (sp) theory about even multiples and how it brings additional clarity that even though I am not sure I fully understand I can credit the possibility and would love a device capable of resolving these kind of resolutions and check myself...


For a digital display device though you should allow the SW to do the scaling and present the display the native panel res as the SW can do this task cleaner (in general) than the HW supplied in most PJ's... SW scaling is not perfect and even multiples helps but you must match the display devices fixed panel (for digital PJ) above all else...
 

· Registered
Joined
·
629 Posts
Discussion Starter · #5 ·
mdv -


The point is, the scaling methods that Photoshop uses, are similar to the methods used by software DVD players or video cards.

We can expect similar (if not the same) results.

And I doubt software DVD players use better scaling methods than Photoshop....

Also, it clearly shows that there is a sweet spot when scaling video from DVD. And that is, any even multiple of 720x480. For example, 1440x480, 1440x960 - even 720x960....

As you can see, 960x540 didn't look too good on the resolution pattern. Where there is fine detail in an image, the same thing will happen.


-Ryan Dinan
 

· Registered
Joined
·
2,961 Posts
mdv,

I think you missed the point.


Wayne(Karnis) has been theorizing that perhaps it is best to scale a DVD in multiples of it's native resolution (720x480) rather than maintaing aspect ratio.


Hence resolution's like 720x480,1440x960,etc. yield a better PQ than resolutions such as 540P,1080i.


Ryan and myself (I won't speak for Karnis) have found this to be true by trying the various resolutions and displaying the 6.75 Mhz resolution test on AVIA.


What Ryan's posting is showing is the effects of a DVD image (720x480) at different scaling resolution's.


Joe
 

· Registered
Joined
·
6,227 Posts
Another thing I forgot to mention in the above post is that having both NTSC and PAL DVD's in a collection would play havoc with this setup requiring multiple resolutions per DVD...


I guess with a CRT with multiple memories this may not be such an issue and powerstrip could do the work but it should not be forgotten...


As a side note to the thread I am often surprised when people seem to be able to clearly see the difference between a 480p DVD and a 540p (from 1080i HDTV) image yet no one thinks that PAL 576p is worth the additional resolution... For me I dont find the audio speed up an issue but maybe this is part of it...
 

· Registered
Joined
·
2,956 Posts
Quote:
Originally posted by mdv
I don't see the point. I doubt that anyone uses Photoshop to scale their DVDs.


Mark
But the software/hardware scaling done by DVD software and video cards will produce similar results. Pull up the anamorphic version of the resolution test pattern in Avia and look at the 6.75Mhz circle. Compare at different resolutions and you'll find that scaling errors can vary quite a bit depending on the resolutions.


I have an HDTV RPTV, and I used to run 1704x960i as my DVD resolution. This a better picture than 540p or 480p timings, and was a 16:9 resolution with no overscan. For the most part I was happy, but I would occasionally see scaling errors in the horizontal dimension (ie vertical lines that aren't as straight as they should be). The 6.75Mhz test circle made the problem quite apparent. I decided to tryu 1440x960i instead. Even though this is not a 16:9 proportion it is an even multiple of the DVD resolution and therefore produces a better picture with no scaling artifacts. Of course, you need DVD software that can handle the non-proportional resolution, but TheaterTek works quite nicely for that.


For digital projectors you don't have as much flexibility since you need to target the native resolution. But the raster-based nature of CRT devices makes varying the horizontal dimension quite easy, and CRT's often support multiple resolutions. It's definitely worth checking out.
 

· Registered
Joined
·
629 Posts
Discussion Starter · #9 ·
Quote:
Originally posted by JKohn



Of course, you need DVD software that can handle the non-proportional resolution, but TheaterTek works quite nicely for that.

Actually, if you simply switch your DVD player to do "full screen", it will work on any DVD software - IF you're running the resolution allready.

Since PowerDVD 4 has the nice "change resolution" feature, I leave my desktop to 960x540p, and have the software change the resolution to 1440x480p for DVD.

The only way this will work on players such as WinDVD, is if your desktop is allready running at 1440x480, and you choose "full screen". It will look perfectly normal, since your TV (if you have a CRT) is drawing the 1440x480 in a 16x9 aspect.


-Ryan
 

· Registered
Joined
·
9,884 Posts
I also like to use a 1440x480 resolution. But in my case it's on a 4:3 set. I made a squashed 1440x480 inside 544p timings and then slightly also squeezed the veritcal on my sets system menu.


This allows me to use almost all the screen in 960x544 for web browsing but still have close to the correct aspect ratio for DVD's without too many scaling artfacts.


- Tom
 

· Registered
Joined
·
93 Posts
This is very intereresting post.


I am doing many tries , and tonight on my Nec XG135LC with a ATI I saw that my usual resolution [email protected] is worst than [email protected]


I am trying to understand wich could be better for Pal resolution, or in general for dvd : I am in doubt if I have use multiples of the dvd res , or what else like using the Bjoyrn Roy system . He uses something like for the 1.85 dvd a res that is exactly the results by horz x vert .

But thing I don't understand is if I want to go in 4x , what I should do ? 1440x 1152 (for Pal ) so perfect multliples of dvd res (720x576 ) but I have not mantein the right aspect ratio , or 2880 (??) x 1152 for getting a number "near" the aspect ratio of 2.35?


Hope to be almost clear in my bad english :)


Fabietto
 

· Registered
Joined
·
2,961 Posts
Here are some timings from my Radeon for 720x480 and 1440x480 that folks may want to try.


Since I use Windows 2000, I can not try 960i.


This set is tweaked for my Mitshubishi so you may have to change them slightly.


1440x480=1440,169,144,159,480,27,6,52,62852,336

720x480=720,62,64,66,480,33,1,49,30807,336



This second one I got from a well known forum member.

1440x480p=1440,123,144,109,480,33,10,40,60942,336


Ryan - how about posting yours?


Joe
 

· Premium Member
Joined
·
2,564 Posts
Quote:
As you can see, 960x540 didn't look too good on the resolution pattern. Where there is fine detail in an image, the same thing will happen.
If you watch 720x480 checkerboards on your HTPC, these tests are definitive.


For other sources, I'm not so sure


Imagine doing the following. Photograph (or generate) a 960x540 image checkerboard. Resample the checkerboard to 720x480 to simulate what this checkerboard would look like if converted to dvd. Resample it again back to 960x540. I imagine neither of the resampled images would look remarkably better than the other. I certainly could be wrong, it would be interesting to see....nyquist theorm comes to mind
 

· Registered
Joined
·
629 Posts
Discussion Starter · #14 ·
Here's my 1440x480p timing - It creates a perfect window with no overscan on a Sony 53HS10.


PowerStrip timing parameters:

1440x480=1440,92,144,108,480,29,6,50,60576,336


Generic timing details for 1440x480:

HFP=92 HSW=144 HBP=108 kHz=34 VFP=29 VSW=6 VBP=50 Hz=60


Linux modeline parameters:

"1440x480" 60.576 1440 1532 1676 1784 480 509 515 565 +hsync +vsync


-Ryan Dinan
 

· Registered
Joined
·
6,653 Posts
For those of you who use Radeon drivers that support interlaced resolutions:


1440x960i=1440,143,144,121,960,54,27,84,62077,344


Adjust the front & back horizontal porches to keep overscan under control (a little is actually ok). Then tweak your aspect ratio controller (YxY or TheaterTek Player, etc) so the 6.75 Mhz circle pattern in AVIA produces evenly spaced lines. Then you will know you've hit the sweet spot.
 

· Registered
Joined
·
554 Posts
Alright, I'm going to go out on a limb and rant a bit about what I see as an ongoing misconception about DVD images and video resolution. The entire DVD industry has been touting that DVD players have 540 horizontal lines of video resolution, but that's pure bunk. Yes, a DVD image containing 720 pixels across can accurately represent 360 black lines interleaved with 360 white lines, thus giving a horizontal resolution of 540 lines, but that's just an artifact of the sampling frequency. It's just like saying an audio CD has high frequency content up to 44.1 KHz. A CD can contain a perfect 44.1 KHz sine wave, but all other frequencies between 22.05 and 44.1 KHz cannot be represented. The Nyquist theory says your sampling rate must be twice the desired signal bandwidth. So a DVD image can only accurately represent 360 horizontal lines of resolution. Any information in a DVD higher than 360 horizontal lines (or 240 vertical lines) is actually fake information that wasn't present in the original image!


If you used 360x240 checkerboard patterns and scaled them to any resolution higher than 720x480 with quality scaling algorithms, you would notice that the black to white transition is still properly maintained at the proper frequency. The scaling algorithms falter with frequencies higher than 360 horizontal lines (and 240 vertical lines) because there's no real information there.


When DVDs are mastered, low-pass filters are supposed to be applied to the signal to restrict it to acceptable frequencies. When these filters are not correctly applied, we get DVDs with aliasing and edge artifacts, i.e. they look "digital". Proper filtering leaves an image looking smooth and natural, but the the video industry has gotten used to having that artificial "snap" to the image, and they don't want to let it go. On small CRT monitors, the artifacts may indeed spruce the image up a bit, but on our large FP systems, the artifacts are annoying, and the scalers just exacerbate them.


Just my two cents :)


Dave
 

· Registered
Joined
·
686 Posts
Phat,

Quote:
As a side note to the thread I am often surprised when people seem to be able to clearly see the difference between a 480p DVD and a 540p (from 1080i HDTV) image yet no one thinks that PAL 576p is worth the additional resolution... For me I dont find the audio speed up an issue but maybe this is part of it...


I've extensively compared Gladiator, R2 Dutch version (Radeon, PowerDVD, can't remember version) @720p against HD Gladiator @720p (HiPix).


I don't know the resolution for the original HD transfer of this movie, but I can confidently state the HD transfer is better on my display (8" CRT on 75" wide screen). Colours are more vibrant, the picture has no apparent noise nor edge enhancement, and there's a small increase in perceived resolution.


I've posted these findings a while back in a thread on HD (can't find it) and someone suggested that perceived resolution with 1080i is closer to something like 720p, not 540p. This seemed reasonable after my findings, though I'm not sure why. OTH, R2 Gladiator really IS close to that HD transfer, and it's better than any R1 I've seen :D


You might also want to check R2 Romeo Must Die, a transfer with excellent resolution, though the movie is not that great. I wish R2 TPMenace looked anything like it :mad:
 

· Registered
Joined
·
629 Posts
Discussion Starter · #20 ·
Quote:
The Nyquist theory says your sampling rate must be twice the desired signal bandwidth. So a DVD image can only accurately represent 360 horizontal lines of resolution. Any information in a DVD higher than 360 horizontal lines (or 240 vertical lines) is actually fake information that wasn't present in the original image!
If a DVD supposedly has 540 lines of horizontal resolution, shouldn't it only be able to display (according to the Nyquist theory) 270 lines? Or are you talking about 360 'pixels'?


So are these resolution patterns in AVIA and VE only good to half of what they state?

I guess I'm not understanding why the detail is "fake", as you put it. It may not be exact, but it's close enough to represent a change. Before I switched to 1440x480p from 960x540p, I could not see individual lines on the 6.75MHz pattern. The fact is, I can see these individual, black and white lines now.


I went ahead and made a grid of 360x240 dots. Here's what I found:

http://members.socket.net/~ryandinan...0/720x48~1.jpg

This is 720x480 upconverted from a grid of 360x240 dots.

http://members.socket.net/~ryandinan...0/1440x4~1.jpg

This is 1440x480 upconverted from a grid of 360x240 dots.

http://members.socket.net/~ryandinan...0/960x54~1.jpg

This is 960x540...

http://members.socket.net/~ryandinan...0/1024x7~1.jpg

And this is 1024x768....


All of these look much better - Obviously, because you have more area to describe a lower resolution source. There are quality differences between these, as you can see. Notice that the 960x540 and the 1024x768 patterns are softened in the vertical direction as well as the horizontal? I still feel that scaling to any resolution that isn't an even multiple of the source, can cause loss of fine detail - "fake" or not :)


-Ryan Dinan
 
1 - 20 of 109 Posts
Status
Not open for further replies.
Top