Avisynth: Are there better filters for FitCD?
 

Hi All,

Like most I usually encode with the default settings created by FitCD,
but are there better filters for resize, smoothing, etc.? :roll: Will others filters
compress the file size as well or better? So far I’ve discovered the following
filters for testing and my issue is there settings? Below is my script:

#
# Compressibility Test
#
LoadPlugin("E:\DVD Backup\2 - DVD2SVCD\MPEG2Dec\MPEG2DEC2.dll")
LoadPlugin("E:\DVD Backup\2 - DVD2SVCD\SimpleResize\Simpleresize.dll")
LoadPlugin("E:\DVD Backup\2 - DVD2SVCD\Unfilter\Unfilter.dll")
LoadPlugin("E:\DVD Backup\2 - DVD2SVCD\Lanczos\Lanczos.dll")
LoadPlugin("E:\DVD Backup\2 - DVD2SVCD\C3D\Convolution3D.dll")
Mpeg2Source("D:\Temp\Movie.d2v")
#SimpleResize(352,240)
#BicubicResize(352,240)
#BilinearResize(352,240)
#LanczosResize(352,240)
#TemporalSmoother(2,2)
#Convolution3D(1,0,0,7,7,3,0)
#UnFilter(5,5)
#Sharpen(.1)
#AddBorders(8,8,8,8 )
###--------------------- Start Of File Size Prediction ----------------------###
IL = Framecount / 100 # interval length in frames
SL = round(Framerate) # sample length in frames
SelectRangeEvery(IL,SL)
## MPEG size = ((Total frames/Framerate)/100) * (MPEG sample file size * .95) ##
###------------------------End File Size Prediction--------------------------###
#
# End of Script

How do they compare?

1) Convolution3D vs TemporalSmoother

2) LanczosResize vs BilinearResize, BicubicRsizse

3) Sharpen vs Unfilter

Which combinations would give the best compression and picture quality? :)

-black prince

Hey black prince,

Too many filters :!: Be careful you don't convolute in the twilight zone :lol:
Back to the question 8) . It would be good to make an identical sample with each of these filters, and compare the size/quality/noise redux, etc. :idea:

-kwag

Hi Kwag,

You'll notice the filters are all commented (#). I uncomment them in
combinations such as, TemporalSmoother and BilinearResize,
or TemporalSmoother and LanczosResize, etc. The picture
quality is hard to judge since they seem close. Is there a way to
determine this imperically rather than subjectively. I once saw
a display of frame(s) magnified to show noise correction on
Doom9, but have had trouble finding the program that produced it.
It showed clearly how certain filters effected picture noise. So far
C3D increases the file size, but appears to be no better than TS(2,2)
which decreases file size in my tests. LanczosResize gives
an added sharpness without using Sharpen(.1) or Unfilter. This reduced
file size. For BicubicResize vs LanczosResize, LanczsosResize has a definate
edge. Going from a lower resolution to a higher one, picture was better.
With maganifed images of frames this would be much easier to see. :|

-black prince

Hi black prince,

I agree with you on the temporal smoother. I get smaller file size with a value of 1,2 without loosing sharpness. Visually, I think there's not that much difference between C3D and TS(1,2). I do see that C3D softens the image more. More tests have to be made :idea:

I know. I was just joking :wink:

-kwag

Kwag,

Your right on target. I found that LanczosResize and TemporalSmoother(2,1)
are great together. You also get the sharpness from Lanczos and file size
is the smallest among my other tests. Picture quality is better for bicubic
low res to high and high to low is same as Bilinear. By the way, hard
coded IL and SL in file size predicting VBR and now I am getting less than
+-1%. Seems Framecount and Framerate may not always be accurate.
Tmpgenc now shows 2400 frames for test with hard coded numbers. 8)

-blac prince

Oops, :roll:

I forgot. Lanczos is a part of MPEG2DEC and needs no loadfile
statement. Just code LanczosResize(width,height) and your set.

-black prince

Hi Kwag,

If your still testing filters, could you add TemporalSoften(1,4,4) to
your list. It appears to have less blurring effects than
TemporalSmoother. :)

Thanx

-black prince