Log in

View Full Version : SAR question


bkman
13th January 2006, 20:16
Hi.

I have looked at quite a few topic dealing with SAR on this forum, but I have never seen a definitive answer to a very basic question of mine; what is the advantage of using a custom SAR over just resizing the image? Is there any benefit to encoding full DVD resolution then setting an SAR that will give 4:3 DAR compared to just sesizing to 640x480? It seems to me that resizing is better because there is less resoltuion to encode, and the decoders that support SAR will just scale with bilinear down to 640x480 anyway.

Can anyone tell me the use of SAR then?

ChronoReverse
13th January 2006, 20:23
I've always used AR correction in the other way... that is, the player scales up.

Basically I keep the vertical resolution and skimp on the horizontal on the assumption that the eye is more sensitive to vertical resolution. Thus bits are saved and can be used for details by the codec.

Then when I playback the video, I use MPC which can utilize pixel shaders to perform a bicubic resize =)

charleski
13th January 2006, 20:28
what is the advantage of using a custom SAR over just resizing the image?
You retain the full resolution of your input. That's the whole point. Using AVC you can quite easily compress a film into the same space as would be used by xvid, but with 40% more resolution. The difference is minimal for 4:3 DARs, but that represents a tiny fraction of the stuff most people encode. The vast majority of inputs are 16:9 for both TV and movies.

bkman
13th January 2006, 20:35
Yes, but what is the use of retaining the full resolution of your input if it is just shrunk down to a lower one by the decoder/player?

As far as I know, PC monitors can't change the shape of their pixels, so the extra resolution will only be lost by the resizing process.

Richard Berg
13th January 2006, 20:38
SAR allows the user to decide where to allocate bits. For example, I often encode bobbed material. Much of the resulting vertical rez isn't really detail (it's interpolated) so I use resize + SAR to throw some away.

In your example, if the clip has lots of fine horizontal detail you don't want to resize away, SAR makes perfect sense.

Richard Berg
13th January 2006, 20:40
As far as I know, PC monitors can't change the shape of their pixels
So? PC monitors also have a lot more than 720 pixels of rez. If you're running at 1600x1200, for instance, resizing from 720->1600 will probably look better than 640->1200.

charleski
13th January 2006, 20:42
Yes, but what is the use of retaining the full resolution of your input if it is just shrunk down to a lower one by the decoder/player?

As far as I know, PC monitors can't change the shape of their pixels, so the extra resolution will only be lost by the resizing process.
????

Where on earth did you get this idea?
When watching on a PC I always scale my video UP since my desktop is 1280x1024. If I scale the video to full width the vertical height is always greater than that of the input.
If you watch the video on an SD TV, you will only be downscaling the video if you watch a 16:9 stream on an old 4:3 set.

bkman
13th January 2006, 20:46
Ok, so, you are saying that a higher source resolution with SAR is only useful when playing at a higher resolution that the source video, because the upscaler has more resoltuion to work with?

I was previously only comparing the benefit at 1X zoom, in which case there wouldn't be any, right?

charleski
13th January 2006, 20:57
I was previously only comparing the benefit at 1X zoom, in which case there wouldn't be any, right?
It's difficult to imagine any situation in which you'd be watching a movie at 1x zoom on a PC monitor outside very specialised circumstances. But yes, if the horizontal resolution was fixed at that of the input, an anamorphically-encoded stream would have to be scaled down vertically. That's the same situation as watching a 16:9 movie on a 4:3 TV.

Richard Berg
13th January 2006, 20:58
I already gave an example where SAR lets me allocate bits more intelligently based on the properties of the clip I send to the encoder. Nobody's screen is 640x336 (a typical resolution for one of my 60fps encodes) so talking about "1X zoom" is irrelevant.

bkman
13th January 2006, 21:03
I was just thinking about 1X zoom as a comparison point, without considering the benefit for scaling to higher resoltions at playback (which you are right is much more common).

Thanks for helping me get my head around it.