BicubicResize / BilinearResize / BlackmanResize / GaussResize / LanczosResize / Lanczos4Resize / PointResize / Spline16Resize / Spline36Resize / Spline64Resize

BicubicResize (clip, int target_width, int target_height, float "b=1./3.", float "c=1./3.", float "src_left", float "src_top", float "src_width", float "src_height")
BilinearResize (clip, int target_width, int target_height, float "src_left", float "src_top", float "src_width", float "src_height")
BlackmanResize (clip, int target_width, int target_height, float "src_left", float "src_top", float "src_width", float "src_height", int "taps=4")
GaussResize (clip, int target_width, int target_height, float "src_left", float "src_top", float "src_width", float "src_height", float "p=30.")
LanczosResize (clip, int target_width, int target_height, float "src_left", float "src_top", float "src_width", float "src_height", int "taps=3")
Lanczos4Resize (clip, int target_width, int target_height, float "src_left", float "src_top", float "src_width", float "src_height")
PointResize (clip, int target_width, int target_height, float "src_left", float "src_top", float "src_width", float "src_height")
Spline16Resize (clip, int target_width, int target_height, float "src_left", float "src_top", float "src_width", float "src_height")
Spline36Resize (clip, int target_width, int target_height, float "src_left", float "src_top", float "src_width", float "src_height")
Spline64Resize (clip, int target_width, int target_height, float "src_left", float "src_top", float "src_width", float "src_height")

From v2.56 you can use offsets (as in Crop) for all resizers:
GaussResize (clip, int target_width, int target_height, float "src_left", float "src_top", float -"src_right", float -"src_top")

For all resizes you can use an expanded syntax which crops before resizing. The same operations are performed as if you put a Crop before the Resize, there can be a slight speed difference.

Note the edge semantics are slightly different, Crop gives a hard absolute boundary, the Resizer filter lobes may extend into the cropped region but not beyond the physical edge of the image.

Use Crop to remove any hard borders or VHS head switching noise, using the Resizer cropping may propagate the noise into the adjacent output pixels. Use the Resizer cropping to maintain accurate edge rendering when excising a part of a complete image.


# which is nearly the same as

Important: AviSynth has completely separate vertical and horizontal resizers. If the input is the same as the output on one axis, that resizer will be skipped. Which one is called first, is determined by which one has the smallest downscale ratio. This is done to preserve maximum quality, so the 2nd resizer has the best possible picture to work with.


The BilinearResize filter rescales the input video frames to an arbitrary new resolution. If you supply the optional source arguments, the result is the same as if you had applied Crop with those arguments to the clip before BilinearResize.

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.)

# Load a video file and resize it to 240x180 (from whatever it was before)

# Load a 720x480 (CCIR601) video and resize it to 352x240 (VCD),
# preserving the correct aspect ratio
AVISource("dv.avi").BilinearResize(352, 240, 8, 0, 704, 480)

# or what is the same
AviSource("dv.avi").BilinearResize(352, 240, 8, 0, -8, -0)

# Extract the upper-right quadrant of a 320x240 video and zoom it
# to fill the whole frame


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 values are 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 a better overall final result.

For the most numerically accurate filter constrain b and c such that they satisfy :-

  b + 2 * c = 1
This gives maximum value for c = 0.5 when b = 0. This is the Catmull-Rom spline. Which is 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 with a TV set sharpness control. Negative values are not allowed for b, use b = 0 for values of c > 0.5.


BlackmanResize is a modification of LanczosResize that has better control of ringing artifacts for high numbers of taps. See LanczosResize for an explanation for the taps argument (default: taps=4). (added in v2.58)


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 uses 4 taps and has similar speed as Lanczos4Resize. (added in v2.56)

LanczosResize / Lanczos4Resize

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.

Lanczos4Resize (added in v2.55) is closely related to LanczosResize (correct name: Lanczos3Resize). The latter uses 2*3=6 lobes and the former 2*4=8 lobes to do the resizing. The result is that Lanczos4Resize produces sharper images. Especially usefull when upsizing a clip.

Lanczos4Resize is a shorthand for LanczosResize(taps=4).

Note: the input argument named taps should really be lobes (actually it is half the number of lobes). When discussing resizers, taps has a different meaning, as described below.

Interpolation Tap Size (taps parameter)

Assuming Taps=4, for upsampling (making the image larger), the filter span 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 is effectivly 8 taps. For 3x downsampling, you get effectivly 3*4=12 taps, and so forth.

Thus the effective number of taps you get for downsampling is the downsampling ratio times the number of filter input taps (thus Tx downsampling and LanczoskResize results in T*k taps), this is rounded up to the next even integer. For upsampling, it's always just k taps. Source: [avsforum post].


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 inferiour quality, or you need the clear pixel drawings.
It is very useful for magnifying small areas of pixels for close examination.


Three Spline based resizers  (added in v2.56/v2.58). The rational for Spline is to be as sharp as possible with less ringing artefacts than LanczosResize produces. Spline16Resize uses sqrt(16)=4 sample points, Spline36Resize uses 6 sample points, etc ... The more sample points that are used, the sharper your clip will be. A comparison between some resizers is given here.

As of today, the authors of AviSynth have no idea how the coefficients as used in the Spline resizers are derived. If you do, please drop us a note.


v2.55 added Lanczos4Resize
v2.56 added Spline16Resize, Spline36Resize, GaussResize and taps parameter in LanczosResize; added offsets in Crop part of xxxResize
v2.58 added BlackmanResize, Spline64Resize

$Date: 2007/04/17 03:29:29 $