|
|
| robber57 |
| Posted: Nov 27 2002, 08:22 PM |
 |
|
Unregistered
|
I've did a search for the topic but nothing returned,although i'm sure i'm not the first to wonder about it...
What is the value "A=-0.6" ?
what stands "A" for?
Aspect ratio? |
|
| fccHandler |
| Posted: Nov 27 2002, 08:47 PM |
|
|
Administrator n00b
Group: Moderators
Posts: 3961
Member No.: 280
Joined: 13-September 02
|
I don't think it stands for anything. 
All I can say is that "A=-1.00" is the sharpest, but whether that's a good thing depends on what you're using the filter for. "A=-0.75" is supposed to produce similar results to Photoshop's bicubic resize.
--------------------
May the FOURCC be with you... |
|
| robber57 |
| Posted: Nov 27 2002, 09:03 PM |
|
|
Unregistered
|
Hm.........
Ok, i mostly use the filter to up-size web video to vcd/Svcd format, so i gues i would be best off with A=0.75...
Thnxxx! |
|
| Sarreq Teryx |
| Posted: Nov 27 2002, 09:09 PM |
|
|
VirtualdubMod Alpha tester
Group: Vdubmod Alpha Testing Team
Posts: 175
Member No.: 41
Joined: 16-July 02
|
| QUOTE (fccHandler @ Nov 27 2002, 02:47 PM) | I don't think it stands for anything.
All I can say is that "A=-1.00" is the sharpest, but whether that's a good thing depends on what you're using the filter for. "A=-0.75" is supposed to produce similar results to Photoshop's bicubic resize. |
I'm guessing, and am probably way off, Aperture?
--------------------
And as I walk through the Valley of the Shadow of Death, Lord, thy balls and shaft, they comfort me, you annoint my head with oil, some salt, a dash of pepper, a sprigg of parsley......Lord?............Lord??? What dost thou intend to do with that fork??? |
|
| phaeron |
| Posted: Nov 28 2002, 03:21 AM |
|
|
Virtualdub Developer
Group: Administrator
Posts: 7773
Member No.: 61
Joined: 30-July 02
|
It controls the stiffness of the two cardinal splines that make up the bicubic filter kernel -- this mainly affects the peaking in the filter around the transition band. The A=-1.0 filter reduces aliasing better than the -0.6 filter, but it also boosts some high frequencies, whereas A=-0.6 is closer to a flat passband response. For the most part, the difference isn't noticeable unless you do a lot of cumulative resizes on the same image, for which A=-0.6 produces the closest result to the original -- but for what you're doing, just choose the one that looks best.
|
|
| fccHandler |
| Posted: Nov 28 2002, 04:28 AM |
|
|
Administrator n00b
Group: Moderators
Posts: 3961
Member No.: 280
Joined: 13-September 02
|
Thanks, phaeron!
This discussion makes me wonder how interesting it might be to rewrite the filter configuration, and let the user specify his own value for "A" in a textbox.
Why are only those three values available? How were they chosen? Are any other values useful? What kinds of effects would more extreme values produce?
Just thinking out loud 
--------------------
May the FOURCC be with you... |
|
| phaeron |
| Posted: Nov 28 2002, 06:54 AM |
|
|
Virtualdub Developer
Group: Administrator
Posts: 7773
Member No.: 61
Joined: 30-July 02
|
A=-1.0 was the original filter equation I found, before I had any idea of the sinc(x) and bicubic spline derivation. I later added A=-0.75 when I found a webpage recommending it as being closer to Photoshop's filter. A=-0.60 got chosen by some work in MATLAB analyzing the frequency response of the filter and searching for the flattest passband -- I forget the ugly expression I used, but presumably the power spectrum is sufficient to choose the "best" filter given that the kernel is symmetric and should be linear phase. I have heard that A=-0.5 produces the best interpolation splines, but the continuous kernel is never used by itself since it's sampled into individual kernels by phase.
Basically, I winged it for a long time until I actually learned digital filter theory (I ain't no EE major).
Modifying the spline tension parameter between A=-0.6 and A=-1.0 has very little visual impact, so I have not been motivated to add additional values in between. It is true that altering that value affects compression ratios with high-compression codecs, but that is better served by allowing the user to manually drop the low-pass filter cutoff or convolve in an additional gaussian blur filter, which is on my to-do list. Truthfully, I think the resize filter performs about as well as a pure low-pass filter based algorithm can do on 8-bit channel video, and any further quality gains will require an adaptive algorithm, like warp sharp.
|
|
| robber57 |
| Posted: Nov 28 2002, 07:56 PM |
|
|
Unregistered
|
Jeez..... 
This is going way over my head... 
I gues i just use the 0.75...
Thnxx guys! |
|
| fccHandler |
| Posted: Nov 28 2002, 08:20 PM |
|
|
Administrator n00b
Group: Moderators
Posts: 3961
Member No.: 280
Joined: 13-September 02
|
 Honestly, it's over my head too. But I love these forums, and I learn something new every time phaeron speaks. I still don't know what "A" stands for (if anything), but "Aperture" sounds good to me.
--------------------
May the FOURCC be with you... |
|
| valja |
| Posted: Nov 28 2002, 10:24 PM |
|
|
Advanced Member
Group: Members
Posts: 179
Member No.: 66
Joined: 1-August 02
|
Thanx, phaeron!
Hopefully I got idea of difference between "bicubic" and "precise bicubic" resize. In case of "cubic resize" function is interpolated in both directions as cubic polynome by four nearest neighbours (four points just defines cubic function). In case of "precise bicubic" there is used cubic spline (and much more math to solve equations). I wasn't able yet to follow your explanation entirely, but at least I know direction where to dig  |
|
