User manual
Creating Projects
41
© 2014 Digital Vision
UM-2014.1-07
format maintaining the source image aspect ratio. This may result in a
cropped or blanked image, depending on the difference between the
source and output resolution and aspect ratio.
Filtering
The filter type selects the mechanism used to transform the pixels in the
source image to the new output. This mechanism involves calculating the
value of an output pixel as an aggregation of contributing source pixels. How
the various pixel contributions are weighed defines the filter.
In order of worst to best (or increasing accuracy) :
No Filtering
Only the source pixel at the center of the projected boundary is
sampled. This method is very prone to visible artefacts such as jagged
image edges.
Bilinear Filtering
The four pixels in the source image that are closest to the center of the
projected boundary are sampled. Their weighting is based on the
relative proximity to the center of the projected pixel boundary. This
produces an accurate colour result but doesn't cut down aliasing as well
as Gaussian does. Box and bilinear will produce very similar results for
small scale reductions, but due to the nature of the software coding
bilinear is generally the fastest of these two options.
Box Filtering
All pixels falling within the projected pixel boundary are sampled and
their weight is proportional to their area that lies within the projected pixel
boundary. Box and bilinear filtering will produce very similar results for
small scale reductions, but box can produce better results for larger
scale reductions.
Gaussian Filtering
All pixels falling within the projected pixel boundary are sampled and
weighted using a gaussian filter kernel. This produces better results
than the box or bilinear filters in so far as it results in less aliasing, but is
slower.
ZOM Filtering
Uses a sinc filter and is the best general filter for up and down-scaling,
providing a sharp result without artefacts. This is the default setting.
Lanczos Filtering
A windowed form of the sinc filter which may give better results in some
cases.