Video filter reference

2:1 reduction	Shrink video to half-size using a non-overlapping 2x2 kernel.
2:1 reduction (high quality)	Shrink video to half-size using an overlapping 3x3 kernel.
blur	Apply a radius-1 Gaussian blur to video.
blur more	Apply a radius-2 Gaussian blur to video.
bob doubler	Interpolate interlaced video to field rate.
box blur	Apply fast approximated blurs to video.
brightness/contrast	Perform fast linear luminance adjustments.
chroma smoother	Re-interpolate color information in an image without affecting brightness.
convert format	Convert video to a different image format.
deinterlace	Split, unsplit, discard, and duplicate fields.
emboss	Apply psuedo-3D edge detection matrices.
field bob	Resamples alternating frames to remove jittering when splitting fields to frames.
field swap	Swaps even and odd fields to fix mistakes.
fill	Fills a rectangle in the video with a solid color.
flip horizontally	Flips video frame horizontally.
flip vertically	Flips video frame vertically.
general convolution	Apply an arbitrary 3x3 filtering matrix.
grayscale	Convert a color video to grayscale.
HSV adjust	Adjust hue, saturation, and value.
invert	Create or undo negatives.
levels	Apply non-linear ramp adjustments.
logo	Plop your very own bug into the frame.
motion blur	Blur video in time to create motion trails.
null transform	Does nothing, but can be used to crop.
resize	Convert video to a different size.
rotate	Rotate video by right angles.
rotate2	Rotate video by arbitrary angles.
sharpen	Make video crisper.
smoother	Adaptively blur a video while trying to preserve edges.
temporal smoother	Adaptively noise-reduce video across frames.
threshold	Convert video to black and while by comparing against a threshold.
TV	Blur or resample video in luma/chroma space.
warp resize	Resize video using an edge-sensitive algorithm.

2:1 reduction

• Shrink video to half-size using a non-overlapping 2x2 kernel.

This filter is equivalent to applying a 2x2 box filter (average 4 pixels) and then doing a point-sampled resize by half. It gives sharper results than 2:1 reduction (high quality), at the cost of some sparkling (aliasing). Where better quality is required, the resize filter should be used in precise bicubic or Lanczos3 mode instead.

Note	This filter is deprecated and will be removed in a later version.

2:1 reduction (high quality)

• Shrink video to half-size using an overlapping 3x3 kernel.

This filter is equivalent to applying a radius-1 Gaussian blur and then doing a point-sampled resize by half. It is marginally better than 2:1 reduction in avoiding aliasing (sparkling), at the cost of a blurrier output. Where better quality is required, the resize filter should be used in precise bicubic or Lanczos3 mode instead.

Note	This filter is deprecated and will be removed in a later version.

blur

• Apply a radius-1 Gaussian blur to video.

blur more

• Apply a radius-2 Gaussian blur to video.

bob doubler

• Interpolate interlaced video to field rate.

This filter converts interlaced video at frame rate to progressive video at field rate. For instance, 30 fps interlaced video is converted to 60 fps without interlacing. Field order and interpolation algorithm can be chosen.

box blur

• Apply fast approximated blurs to video.

Box blur is so named because it uses "box" filters -- it averages blocks of pixels together. Two passes gives a triangle filter, three gives a quadratic, and four gives a cubic. The advantage of box blur is that it can do very large blurs very quickly, such as radius-50 cubic blur. The main disadvantage is that it cannot do small or intermediate (radius-5.2) blurs.

brightness/contrast

• Perform fast linear luminance adjustments.

Brightness/contrast provides quick and dirty adjustments -- one of the side effects that saturation is affected by contrast, and hues can shift if clamping at white or black occurs. Consider using levels or HSV adjust for non-subtle adjustments.

chroma smoother

• Re-interpolate color information in an image without affecting brightness.

Video compression algorithms often store color (chroma) information at a lower resolution than grayscale brightness information, since the eye is less sensitive to detail in color; however, some video codecs don't interpolate chroma well on decompression, resulting in blockiness around sharp color transitions. The chroma smoother filter attempts to apply a blur solely in the color channels to correct this situation.

Each chroma option in chroma smoother's configuration corresponds to a blur that is tuned for a particular type of chroma subsampling; assuming that the video decoder did not interpolate chroma at all, choosing the correct setting will produce a result approximating bilinear interpolation. Of course, stronger or weaker options can be picked to taste.

convert format

• Convert video to a different image format.

Use this filter to force video to be converted to a different image format. This is mainly useful for diagnostic purposes, although it is also useful in capture mode to force a specific output format.

deinterlace

• Split, unsplit, discard, and duplicate fields.

This filter applies quick-and-dirty field operations -- discarding one field is a brutal, but quick-and-dirty way to remove combing from the image. For better quality an adaptive deinterlacer should be used instead. The fold and unfold modes allow you to temporarily split a video into a double-width video, with even fields on the left and odd fields on the right. This essentially allows any filter to be applied on a field basis rather than a frame basis, avoiding blurring between fields which shows up as ghosting in interlaced output.

emboss

• Apply psuedo-3D edge detection matrices.

Emboss gives a 3D-like effect by "lighting" the video from a particular direction. This highlights edges in the image. It can also be used to amplify a video in order to look for low-level noise.

All forms of the emboss filter can also be done manually by the general convolution filter.

field bob

• Resamples alternating frames to remove jittering when splitting fields to frames.

field swap

• Swaps even and odd fields to fix mistakes.

fill

• Fills a rectangle in the video with a solid color.

If you need a translucent (alpha-blended) fill, use the logo filter with a solid color as the logo.

flip horizontally

• Flips video frame horizontally.

flip vertically

• Flips video frame vertically.

general convolution

• Apply an arbitrary 3x3 filtering matrix.

A convolution filter is a filter that uses a local area of pixels to compute a new pixel. It can be thought of as either taking a weighted sum of an area of source pixels, or adding a weighted area to the output scaled by the source pixel (the two are equivalent). Filters that can be performed using convolution include blurs, sharpens, and edge detectors.

The general convolution filter allows a 3x3 filter to be specified. Each value in the filter is specified as a value from a scale to 0-256, where 256 means a value of 1.0 (yes, this is a programmerism). Values can be specified outside this somewhat, including negative numbers, although large values like 10,000 are inadvisable. Each value then specifies either the amount of source to include in the result, or the amount of result to produce from the center source pixel (again, depending on your perspective).

The clipping option should almost always be enabled; if it is not, the result will be incorrect when the result of the filter is below black or above white.

Bias is a value to add to the result, again in the range 0-255. It is useful when both positive and negative values need to be visualized, such as error values. It acts much like a brightness adjustment.

Some hints regarding design of convolution filters:

• The sum of all of the values in the matrix is the gain of the filter, which determines the change in contrast produced by the filter. A sum of 256 results in unity gain, i.e. no change. Higher sums amplify the image and lower sums mute it. A zero sum completely removes the image, and a negative sum inverts it.
• Filter matrices that consist of all positive elements are blurs. For instance, here is a blur matrix:

  28 28 28
  28 32 28
  28 28 28
  

• Matrices that aren't symmetrical — those that are lopsided in a direction — will have edge detection effects.
• A bright pixel in the image will take on the shape of the filter matrix.

grayscale

• Convert a color video to grayscale.

The formula used for conversion is Y = 0.211R + 0.715G + 0.074B.

HSV adjust

• Adjust hue, saturation, and value.

You can rotate hues, control the strength of color (saturation), or modify brightness (value) using this filter. HSV adjust does not allow value adjustments to affect hue -- if a pixel becomes too bright or dark, its saturation is reduced accordingly to avoid clamping. This prevents orange, for instance, from becoming hot yellow when the image is amplified.

invert

• Create or undo negatives.

levels

• Apply non-linear ramp adjustments.

Levels does contrast, brightness, and gamma adjustments. It actually consists of three steps:

1. Scale input range up to full range. The black level and white levels are the first and third parameters on the top; anything lower than the black level becomes black, and anything above the white level becomes white.
2. Apply gamma correction (middle top parameter). You can either type in the gamma correction power, or you can move the slider, which controls which level in the source image becomes middle gray (50%). Black and white always stay black and white through this correction step.
3. Scale to output range (bottom parameters) -- the image levels are finally scaled down to the black and white levels indicated here.

You also have the option of working in luma (Y) space. This causes levels to work in brightness (luma) only, leaving color information alone. This preserves hue and saturation in the image, except for possible clipping at black and white.

Note	Levels always works with full scale levels -- black and white are 0 and 255, not 16 and 235 as in YCbCr space.

logo

• Plop your very own bug into the frame.

Enable per-pixel blending using the alpha channel

An image can be augmented using an additional monochrome map called an alpha channel. This map contains a single value per pixel indicating how opaque or translucent that pixel should be; it allows logos to be created that have shaded areas. Most photographic image editing packages have options to edit an alpha, mask, or opacity layer on an image for this purpose. For use with this filter, it is best to save such images in TARGA (.tga) format with 32-bit pixels.

Use premultiplied alpha

Affects the interpretation of the alpha channel. Normally alpha channels simply control blending of an image, but when creating an image from scratch that has soft edges, creating such an alpha channel may be difficult. Premultiplied alpha refers to a method of blending where the image itself is expected to be pre-darkened where it is not fully opaque, to the point of being black where alpha indicates transparent areas. In a painting package, you may find this more convenient to create, as you can simply draw on top of "transparent black."

Use gray channel of secondary image for alpha channel

When it is inconvenient to create a single image with an alpha channel, enabling this option will allow the alpha to be supplied using the brightness of a second image. This image should have the same size as the main logo image, although it need not be of the same format.

Opacity

Global translucency control; 0% will hide the logo, 100% shows the logo at full opacity, and intermediate values will fade it out somewhat. When per-pixel blending is enabled, this option will scale down all of the alpha values.

Justification, X/Y offset

Specifies the position of the image. If justification is something other than TL (top-left), the positioning will be relative to another point within the image, which is useful for consistently placing logos in the bottom-right corner when working with videos of different sizes.

motion blur

• Blur video in time to create motion trails.

null transform

• Does nothing, but can be used to crop.

resize

• Convert video to a different size.

When resampling video from one size to another, pixels in the new frame size don't necessarily map exactly to pixels in the source. A resampling filter is used to compute the intermediate pixels from the pixels in the source. VirtualDub's resize filter gives you a number of choices for the resampling filter:

Nearest neighbor (point sampling)

Choose the nearest source pixel. This results in the crispest video, but has sparkling and "chunkiness" problems. It is the fastest resampling mode and is useful for previews.

Bilinear (triangle interpolation filter)

Compute the desired pixel by linearly averaging the closest four source pixels. This gives a considerably better result than nearest neighbor, but results in a lot of blurring and gives diamond-shaped artifacts when enlarging. This is the resampling mode for most 3D texture mappers and interpolating video hardware overlays. High-ratio shrink operations (<~60% or so) will give aliasing with this mode and for those the precise bilinear mode should be used instead.

Bicubic (cubic spline interpolation filter)

Compute the desired pixel by fitting cubic spline curves to the closest 16 source pixels. This gives a sharper result than the bilinear filter, although when enlarging it results in a slight halo (ringing) around edges. High-ratio shrink operations (<~60% or so) will give aliasing with this mode and for those the precise bicubic mode should be used instead.

Precise bilinear (triangle decimation filter)

Compute the desired pixel by applying a triangle filter to the closest N source pixels, where N=4 for enlarging and N>4 for shrinking. This mode is the same as bilinear for enlargement but gives better results when shrinking.

Precise bicubic (cubic spline decimation filter)

Compute the desired pixel by applying a triangle filter to the closest N source pixels, where N=16 for enlarging and N>16 for shrinking. This mode is the same as bicubic for enlargement but gives better results when shrinking. Three different modes are given, A=-1.0, A=-.75, and A=-0.6. These vary the "stiffness" of the cubic spline and control the peaking of the filter, which perceptually alters the sharpness of the output. A=-0.6 gives the most consistent results mathematically, but the other modes may produce more visually pleasing results.

Lanczos3 (three-lobed decimation filter)

Compute the desired pixel by applying a three-lobed sinc filter to the closest N source pixels, where N=64 for enlarging and N>64 for shrinking. This produces slightly better results than the precise bicubic mode, at the expense of slower speed and more haloing (ringing). However, for a single pass the difference is very small and you should consider using precise bicubic instead.

The resize filter uses OpenGL conventions for pixel mapping, mapping the entire source texture onto the entire destination rect. Areas in the image are proportionally enlarged or shrunk with the picture frame: enlarging the frame 2x results in all features in the image being twice as large. This means that high enlargement ratios will result in some duplicated pixels around the border of the destination image (U/V clamping).

Note

The new width and new height fields do not have to be integers — they can be fractional. VirtualDub will resize the video to that fractional size and then extend the borders to hit the next highest integer size in pixels. The means that exact aspect ratios can be maintained in the image content of a video even though the frame size is slightly off due to integral rounding.

rotate

• Rotate video by right angles.

rotate2

• Rotate video by arbitrary angles.

While rotate2 can be used for right-angle rotations, that job should be left to rotate, which can do them faster and more accurately.

rotate2 is equivalent to a 1:1 texture-mapped quad, rotated around the center point of the quad by the designated angle, with linear or cubic spline interpolation filters applied to the texture mapping.

sharpen

• Make video crisper.

Sharpen cannot distinguish between artifacts, noise, and edges -- so sharpening edges with this filter will also make blocking and noise worse. Thus, there is a limit to how much video can be improved with sharpening, particularly highly-compressed or noisy video.

smoother

• Adaptively blur a video while trying to preserve edges.

temporal smoother

• Adaptively noise-reduce video across frames.

A temporal filter is one that works along time -- temporal smoother doesn't compare pixels within a frame, but across a 7-frame window. It is effective at reducing noise, but for best effect, it should be combined with a filter that works within a frame (spatial filter). Raising the power of the filter increases the amount of noise reduction but can also result in smearing of details when motion occurs as well as sparkling during transitions. It works best with a setting of 3-5.

threshold

• Convert video to black and while by comparing against a threshold.

TV

• Blur or resample video in luma/chroma space.

Due to the way that analog video encoding works, the color information in the video (chroma) typically has lower resolution and more noise than the brightness information (luma). This problem is especially severe with videotape. The TV filter can reduce chroma noise by averaging only the chroma, while leaving the luma alone. This can reduce color flickering in a video, but should be used carefully as too much chroma blurring will lead to bleeding.

warp resize

• Resize video using an edge-sensitive algorithm.

Warp resize attempts to resize video more intelligently by detecting edges in the video and warping areas to sharpen edges. It works better with computer-generated video or animations than with natural video.