The LightWave 3D renderer includes a full implementation of reconstruction filters. These filters determine how the rendering data is combined into a final image at a sub-pixel level in order to produce a final image. Reconstruction filters have been a subject of research over many years and the names of these filters are often derived from the research that led to their development. Reconstruction filters are not set on the Camera Properties panel, but rather per buffer in the Buffer tab of the Render Properties panel.
Different buffers can have different filtering applied but beware of mixing different filters on buffers for the same render since it will be difficult and perhaps impossible to reconstruct a final image.
LightWave filters are outlined below:
This is the most common form of filtering, and the method that most traditional applications use to reconstruct an image from raw rendering samples. These samples are simply places in a pixel “box” and averaged. Although this technique is fast, it can exhibit significant artifacts on motion and when there is fine image detail. From a signal processing point of view, this is a very poor technique for reconstructing an image from the raw data that comprises it. This mode is very close to the traditional LightWave modes.
Based on Box, it uses a 6x6 matrix based on the Filter Radius size. Anything within the filter radius is given full importance. Anything outside is cut off.
Based on a cone with filtering taking place on pixels in a 6x6 matrix.
Gaussian filter based image construction takes the samples that compose the image and builds the final pixel data by weighting their contributions based on a Gaussian kernel of approximately one pixel in size. This technique typically performs quite well, although images tend to have a “soft” look. In practice this mode can be very valuable for video output where it can help hide some of the artifacts introduced by fielded or reduce bandwidth content.
Mitchell filtering is a technique that is now very popular and was suggested as an alternative to Lanczos filtering (see below). It does not suffer many of the ringing artifacts of Lanczos filtering and generally is a very good starting point for most situations.
Lanczos filtering is arguably the technique that yields the closest to the perfect results for image reconstruction. This technique is based on a windowed sinc function (a sinc function being the optimal infinite image reconstruction filter). Unfortunately in practice Lanczos image reconstruction tends to produce overly emphasized edges and “ringing” in high contrast areas of the image.
Box, Circle, Triangle and Gaussian may provide better results for moving images, whereas Mitchell and LanczosSinc may be more suited to stills due to their tendency to sharpen the image.