• For all resizers you can use an expanded syntax which crops before resizing. The same operations are performed as if you put a Crop before the Resize.
  

• AviSynth has completely separate vertical and horizontal resizers. If input is the same as output on one axis, that resizer will be skipped. The resizer with the smallest downscale ratio is called first. This is done to preserve maximum quality, so the second resizer has the best possible picture to work with. Data storing will have an impact on what mods should be used for sizes when resizing and cropping; see the Crop page.

Crop(10, 10, 200, 300).BilinearResize(100, 150)
#... CL  CT  CW   CH                   RW   RH

which is nearly the same as:

BilinearResize(100, 150, 10, 10, 200, 300)
#...            RW   RH  CL  CT   CW   CH
#("CL"=crop left" etc, "RW"="resize width" etc)

AviSource("video.avi").BilinearResize(240,180)

AviSource("dv.avi").BilinearResize(352, 240, 8, 0, 704, 480)

which is the same as:

AviSource("dv.avi").BilinearResize(352, 240, 8, 0, -8, -0)

BilinearResize(320, 240, 160, 0, 160, 120)
#...            RW   RH   CL  CT  CW   CH
#("CL"=crop left" etc, "RW"="resize width" etc)

BicubicResize(clip clip, int target_width, int target_height [, float b, float c ] )
BicubicResize(clip clip, int target_width, int target_height [, float b, float c,
     float src_left, float src_top, float src_width, float src_height ] )

BilinearResize uses standard bilinear filtering and is almost identical to VirtualDub's "precise bilinear" resizing option. It's only "almost" because VirtualDub's filter seems to get the scaling factor slightly wrong, with the result that pixels at the top and right of the image get either clipped or duplicated. (This error is noticeable when expanding the frame size by a factor or two or more, but insignificant otherwise, so I wouldn't worry too much about it.)

(Does anyone know the origin of this statement? The premise is not correct, as with standard mappings you must have duplicated or mirrored pixels around the border when enlarging. Looking at the sources and comparing outputs, VirtualDub 1.6 and Avisynth 2.5 appear to generate the same mappings. I wonder if this discrepancy has already been resolved. ? Avery)

BilinearResize(clip clip, int target_width, int target_height )
BilinearResize(clip clip, int target_width, int target_height [,
     float src_left, float src_top, float src_width, float src_height ] )

BicubicResize is similar to BilinearResize, except that instead of a linear filtering function it uses the Mitchell-Netravali two-part cubic. The parameters b and c can be used to adjust the properties of the cubic, they are sometimes referred to as "blurring" and "ringing" respectively.

With b = 0 and c = 0.75 the filter is exactly the same as VirtualDub's "precise bicubic" and the results are identical except for the VirtualDub scaling problem mentioned above. The default is b = 1./3. and c = 1./3., which were the values recommended by Mitchell and Netravali as yielding the most visually pleasing results in subjective tests of human beings. Larger values of b and c can produce interesting op-art effects--for example, try b = 0 and c = -5.

If you are magnifying your video, you will get much better-looking results with BicubicResize than with BilinearResize. However, if you are shrinking it, you are probably just as well off, or even better off, with BilinearResize. Although VirtualDub's bicubic filter does produce better-looking images than its bilinear filter, this is mainly because the bicubic filter sharpens the image, not because it samples it better. Sharp images are nice to look at--until you try to compress them, at which point they turn nasty on you very quickly. The BicubicResize default doesn't sharpen nearly as much as VirtualDub's bicubic, but it still sharpens more than the bilinear. If you plan to encode your video at a low bitrate, I wouldn't be at all surprised if BilinearResize yields better quality.

You have to set "b + 2 * c = 1" for the numerically most accurate filter. This gives for b = 0 the maximum value for c = 0.5, which is the Catmull-Rom spline and a good suggestion for sharpness.

From c>0.6 the filter starts to "ring". You won't get real sharpness, what you'll get is crispening like on a TV set. Negative values for b (although allowed) give undesirable results, so use b = 0 for values of c > 0.5.

BlackmanResize(clip clip, int target_width, int target_height )
BlackmanResize(clip clip, int target_width, int target_height [,
     float src_left, float src_top, float src_width, float src_height, int taps ] )

BlackmanResize is a modification of LanczosResize that has better control of ringing artifacts for high numbers of taps. See LanczosResize for an explanation of the taps argument (default: taps=4, 1<=taps<=100).

GaussResize(clip clip, int target_width, int target_height )
GaussResize(clip clip, int target_width, int target_height [,
     float src_left, float src_top, float src_width, float src_height, float p ] )

GaussResize uses a gaussian resizer with adjustable sharpness parameter p (default 30). p has a range from about 1 to 100, with 1 being very blurry and 100 being very sharp. GaussResize has similar speed as Lanczos4Resize. (added in v2.56)

LanczosResize(clip clip, int target_width, int target_height )
LanczosResize(clip clip, int target_width, int target_height [,
     float src_left, float src_top, float src_width, float src_height, int taps ] )

Lanczos4Resize(clip clip, int target_width, int target_height )
Lanczos4Resize(clip clip, int target_width, int target_height [,
     float src_left, float src_top, float src_width, float src_height ] )

LanczosResize is an alternative to BicubicResize with high values of c about 0.6 ... 0.75 which produces quite strong sharpening. It usually offers better quality (fewer artifacts) and a sharp image.

Lanczos was created for AviSynth because it retained so much detail, more so even than BicubicResize(x,y,0,0.75). As you might know, the more detail a frame has, the more difficult it is to compress it. This means that Lanczos is NOT suited for low bitrate video, the various Bicubic flavours are much better for this. If however you have enough bitrate then using Lanczos will give you a better picture, but in general I do not recommend using it for 1 CD rips because the bitrate is usually too low (there are exceptions of course).

The input parameter taps (default 3, 1<=taps<=100) is equal to the number of lobes (ignoring mirroring around the origin).

Lanczos4Resize is a short hand for LanczosResize(taps=4). It produces sharper images than LanczosResize with the default taps=3, especially useful when upsizing a clip.

Warning: the input argument named taps should really be lobes. When discussing resizers, taps has a different meaning, as described below (the first paragraph concerns LanczosResize(taps=2)):

"For upsampling (making the image larger), the filter is sized such that the entire equation falls across 4 input samples, making it a 4-tap filter. It doesn't matter how big the output image is going to be - it's still just 4 taps. For downsampling (making the image smaller), the equation is sized so it will fall across 4 *destination* samples, which obviously are spaced at wider intervals than the source samples. So for downsampling by a factor of 2 (making the image half as big), the filter covers 2*4=8 input samples, and thus 8 taps. For 3x downsampling, you need 3*4=12 taps, and so forth.

Thus the total number of taps you need for downsampling is the downsampling ratio times the number of lobes (thus Tx downsampling and LanczoskResize results in T*2*k taps). And practically, one needs to round that up to the next even integer. For upsampling, it's always 2*k taps." Source: avsforum post (mirror).

LanczosResize(clip clip, int target_width, int target_height, int taps)

PointResize(clip clip, int target_width, int target_height )
PointResize(clip clip, int target_width, int target_height [,
     float src_left, float src_top, float src_width, float src_height ] )

PointResize is the simplest resizer possible. It uses a Point Sampler or Nearest Neighbour algorithm, which usually results in a very blocky image. So in general this filter should only be used, if you intend to have inferior quality, or you need the clear pixel drawings. Useful for magnifying small areas for examination.

Spline16Resize(clip clip, int target_width, int target_height )
Spline16Resize(clip clip, int target_width, int target_height [,
     float src_left, float src_top, float src_width, float src_height ] )

Spline36Resize(clip clip, int target_width, int target_height )
Spline36Resize(clip clip, int target_width, int target_height [,
     float src_left, float src_top, float src_width, float src_height ] )

Spline64Resize(clip clip, int target_width, int target_height )
Spline64Resize(clip clip, int target_width, int target_height [,
     float src_left, float src_top, float src_width, float src_height ] )

Spline16Resize, Spline36Resize and Spline64Resize are three Spline based resizers. They are the (cubic) spline based resizers from Panorama tools that fit a spline through the sample points and then derives the filter kernel from the resulting blending polynomials. See this thread for the details.

The rationale for Spline is to be as sharp as possible with less ringing artifacts as LanczosResize produces. Spline16Resize uses sqrt(16)=4 sample points, Spline36Resize uses 6 sample points, etc ... The more sample points is used, the sharper your clip will get. A comparison between several resizers is given here.

SincResize(clip clip, int target_width, int target_height )
SincResize(clip clip, int target_width, int target_height [,
     float src_left, float src_top, float src_width, float src_height, int taps ] )

SincResize is added in v2.6 and it uses the truncated sinc function as resizer. See LanczosResize for an explanation of the taps argument (default: taps=4; 1<=taps<=20).