Resize Filters : Compare/Questions

[anomiERIC]

I did some comp tests with Gkot 0.28.5.2 (got the same numbers on the first movie with .29 alpha 3 so i didnt bother doing the rest) using the Neutral Bicubic and Lanczos resize filters.

Ive pretty much always used Neutral for my resizes, but ive been getting the impression from people that the Lanczos is better. And every time i try the two Neutral always gives me better numbers. So I was wondering just what the difference between the two is. Since Gknot now defaults to Lanczos and all i've been curious.

Here are the movies I tested. I always encode the audio first before i do the comp test to get more accurate numbers and select the file on the bitrate tab. These are all done in DivX 5.0.5 with only B-Frames, no GMC or Psychovisual or the likes. And final file size set to 699 also using default Smart Crop All. Cant think of anything else that might be pertinent information. If there is let me know. Any explanations/thoughts/comments would be appreciated. Im just always looking for ways to get better quality rips =)

Hot Shots 16-9 Force Filmed.d2v

Lanczos @ 512 = .307 81.5%
Neutral @ 512 = .307 87.9%

The Flamingo Kid 4-3 Force Filmed.d2v

Lanczos @ 512 = .189 73.8%
Neutral @ 512 = .189 77.6%

Good Morning Vietnam 4-3 Force Filmed.d2v

Lanczos @ 512 = .151 43.0%
Neutral @ 512 = .151 46.6%

The iceman cometh 4-3 interlaced.d2v

Lanczos @ 512 = .151 50.5%
Neutral @ 512 = .151 53.7%

Using Decomb.dll 1.5.1 w/
Telecide()
Decimate(cycle=5)
only on Iceman Cometh since none of the others were interlaced.

[anomiERIC]

Almost forgot, the 512 is the resolution I did the tests at. I like to get as close to 512 as I can or more =)

[midi]

Isn't the whole point of Lanczos that it preserves more detail/sharpness than bicubic? Of course it's going to lose to neutral in a compressibility test.

Compare it to various sharp bicubic resizes and compare quality if anything. It looks like you still use CDs, though, so your bitrate is too low to use it in practice anyway IMHO.

[Selur]

Yup, lanczos was ment to preserve more details, though I personally would guess that the difference between lanczos and the neutral bicubic will get lost through the resizing down to 512. (I suspect that your source is full pal/ntsc.)

[anomiERIC]

Ok so if it preserves more detail then I would assume a lower compressibility check number will give the same if not better quality as the number from neutral? I understand 'better quality' is a relative term also.

Im resizing down to 512 because its going on 1 CD.

[bond]

here are some test values i made (only take the values to put the filters into relation between each others!)

Matrix - CompCheck 2% (3934 Frames) - XviD_030624-Koepi_hvs-good_VHQ4_CM_QPel_CO_B2-150-75-0 - Virtualdubmod - mpeg2dec3!


No ResizeFilter 32:42 2,00fps 43,6
LanczosResize 18:39 3,52fps 49,4
SharpBicubic 18:30 3,54fps 50,0
NeutralBicubic 18:25 3,56fps 51,9
SimpleResize 17:52 3,67fps 52,6
SoftBicubic 18:09 3,61fps 56,8
BilinearResize 17:59 3,65fps 58,1

[dTb]

The only way to tell what will give you the best quality is to test yourself. It might seem logical that the detail preserved by Lanczos would result in a better quality final encode but at low bitrates the codec is sacrificing detail anyway. Do some test encodes at the same bitrate with both resize filters and compare the final result to see what you prefer.