Best Resize Method

[Snollygoster]

I was wondering now that we have both simple resize and in-codec (DivX Pro 5.02) resize, which method is the best? DivX networks propose to use in codec resize.

In this case can we take the values off Gordian Knot and just put them in the codec's resize tab?

[raistlin2k]

Just finished a benchmark on in-built resizing.

First of all, it slows down the transcoding.

So stay with your common resizing method.

Just one remark: The inbuilt deinterlacer (the one of the pro-edition, small deinterlacer in standard is crap) works only perfect if you let divx do the resize. LOGICAL, this must apply before resizing, otherwise the lines ca't get separated correctly into up &down-frames.

Raist

[Snollygoster]

OK, so we stick to simple resize. Thank you.

[Acaila]

OK, so we stick to simple resize.
If you want speed, yes. If you want quality, no.

[Snollygoster]

Of course I want quality. So Acaila, internal resize is better for quality? What about deinterlacing and IVTC?

[N_F]

I think Acaila means that Sharp Bicubic (and perhaps Neutral, depending on who you ask) will give you "better" picture quality, at the cost of speed (and compressability).

[kilg0r3]

NOT at the cost of compressability. files resized with simple resizer tend to be less compressable. search the forum for similar statements.

[Awatef]

@ Snollygoster
Do all deinterlacing, ivtc, crop & resizing jobs externally.
For Resizing I use BicubicResize Neutral (0.5)
For Deinterlacing I use FieldDeinterlace() of Decomb.
For IVTC I use the newest IVTC 2.2 with the following command:
ivtc(44,11,110)
or if you have especially hard to IVTC video:
ivtc(44,11,110)
FieldDeinterlace()

[N_F]

Originally posted by kilg0r3
NOT at the cost of compressability. files resized with simple resizer tend to be less compressable. search the forum for similar statements.

Perhaps it's you who should search the forum ;)

http://forum.doom9.org/showthread.php?s=&threadid=25659

[temporance]

Hello Raist,

Originally posted by raistlin2k
Just finished a benchmark on in-built resizing.

First of all, it slows down the transcoding.


This is not my experience - for me every built-in resize algo is very very fast. Much faster than external resizing.

Can you post with all of your settings and CPU type (p4/AMD) ?

[theReal]

I did a quick test with a few minutes of DVD material recently when Simple Resize was included in Gknot for the first time. The results were, SimpleResize gets the same compressibility as Neutral Bicubic, but it looks almost like VDub's Precise bilinear. VDub's Precise bilinear, as a fact makes the file much more compressible, though (the compress. test values were about like this: Simple Resize and Neutral Bicubic around 51%, Soft Bicubic around 58%, Precise Bilinear around 60%).

That makes it clear for me: if I want quality, I use Neutral Bicubic, even it is a little (but not too much) slower than SimpleResize. If I need the best compressibility, I use VDub's Precise Bilinear and go through the hassles of getting a much lower encoding speed. Simple Resize somehow doesn't fit into my encoding strategies, I'm not going to use it just because it is a little faster.

[Awatef]

Just a word about SimpleResize: this filter makes too much jaggies, it isn't recommended anyway.
I use it in only one case: resizing in one direction, for example from 720x480 to 480x480. In such a case it doesn't make jaggies.

[kilg0r3]

@n_f drat! only read the first post and then confused the numbers

[DJ Alik]

I would like to revive this topic. I have noticed that Doom9 doesn't even mention Simple Resize in his Gordian Knot guide is the guide outdated or does he not recomend it at all?

I used to use neutral bicubic all time but just now tried to resize a movie using SimpleResize. Should I use SimpleResize if I just want best quality and the speed of the encoding doesn't matter?

I used the latest version of SimpleResize 0.3.2.0

[theReal]

Should I use SimpleResize if I just want best quality and the speed of the encoding doesn't matter? I wouldn't use it then. It's a bilinear resizing algorithm and it doesn't provide such a sharp picture and true colors like Bicubic Resize (or even better Lanczos Resize).

[OvERaCiD23]

As stated earlier, Bilinear/SimpleResize should only be used when resizing in one direction. Otherwise, you do get jagged edges. These also are very soft resizing filters, meaning the picture will not have the crispness one might desire. The whole thing with encoding is compressibility. You want the sharpest picture with the best compressiblity. You could use Lanczos for everything, but the codec would put too many bits for the sharp edges that compression artifcats would appear elsewhere in the frame. Neutrral Bicubic is the most used resizing method, as it provides a good level of sharpness as well as compressiblity.

Don't use DivX5's internal resizing/deinterlace/filtering! It's not even an argument that AviSynth will produce better results.