Spatial Preprocessing - A Practical Guide to Video and Audio Compression - page 557

Four adjacent pixels of different values Nearest neighbor interpolated

0% 0% 0%

21% 59% 59% 100% 100%

21% 59% 100%

Pixel color values Pixel color values

Figure 34-7 Nearest neighbor.

In all of these cases, some form of interpolation is necessary. The spatial interpolation
must be processed separately from the temporal interpolation. A variety of different algo-
rithms have been developed to deal with spatial interpolations.

34.4.3 Nearest-Neighbor Interpolation

Taking into account the relative positions in the raster of each line, the nearest line to the
correct position is simply duplicated. Figure 34-7 shows how a 4-pixel-wide area is scaled
to a 7-pixel-wide area.

34.4.4 Linear Interpolation

This technique refines the nearest-neighbor approach. It calculates a proportional differ-
ence between two adjacent pixels and selects a value that is a blend of the two. This is not
a discrete computation just based on two pixels; it is a curve-fitting exercise based on the
rate of change of the pixel values. A line is fitted to the values on the input grid and
the output values are calculated proportionally according to the output grid positions.
Figure 34-8 shows the same group of 4 pixels scaled using a linear interpolator.

Four adjacent pixels of different values Linear interpolated

7%

0% 0%

20% 40% 62% 86% 100%

21% 59% 100%

Pixel color values Pixel color values

Figure 34-8 Linear interpolation.