Smartek UCC User Manual page 138

In comparison with the bilinear interpolation, the HQ Linear interpolation method has the modest increase
in computational complexity. However, the main advantage of this method is that it generates significant
higher quality color images with greatly reduced edge artifacts. Therefore HQ Linear interpolation is in the
CameraSuite SDK used as the standard demosaicing algorithm.
Pixel Group Interpolation
Pixel Grouping is another interpolation method considering pixel values in a 5x5 neighborhood to calculate
missing color values. It basically works in two phases; first it computes all the unknown green values, and
then it uses the input data along with the green values computed in the first phase, to compute all the
missing red and blue values. The main principle is to determine the gradients in the four directions from
the current processed pixel and select the value with the smallest one for final calculation. The smallest
gradient value is chosen to reduce edge artifacts due to the fact that a higher gradient value is an indication
for edge transition.
In comparison with the bilinear and HQ Linear interpolation methods, Pixel Grouping is the most memory
and computational intensive algorithm. However, the result color image is at very high quality with very
little edge artifacts, especially for scenes with large areas of uniform colors that are separated by clear
boundaries.
Colorized Output
The Colorized algorithm is not doing any interpolation. It simply creates an intensity Bayer RGB color image
by setting the missing color values to zero. The intensity value of the current pixel remains unchanged.
Restrictions at the Image Borders
Nearly all interpolation methods have problems at the borders of the image. Depending on the size of the
filter used (3x3, 5x5, ...), one or more neighbors in each direction are needed for interpolation; at the borders
at least one direction is not available, like illustrated in Figure 108 and Figure 109 for Bilinear and HQ Linear .
Figure 108: Bilinear algorithm border
The ImageProcAPI therefore provides four approaches:
leave border pixels as original (RAW)
cut off border pixels
fill border pixels with a solid color
interpolate border pixels with a specific demosaicing algorithm
Figure 109: HQ Linear algorithm border
132
User Manual - twentynine
SMARTEK Vision | User Manual - twentynine | Doc. v1.0.2