The functions described in this section generate and reconstruct Gaussian and Laplacian image pyramids.
Figure “Three-level Gaussian and Laplacian Pyramid” shows the basics of creating Gaussian or Laplacian pyramids. The original image G0 is convolved with a Gaussian, then down-sampled to get the reduced image G1. This process can be continued as far as desired or until the image size is one pixel.
The Laplacian pyramid can be built from a Gaussian pyramid as follows: the Laplacian level k, that is, Lk can be built by up-sampling the lower level image Gk+1. Convolving the image with a Gaussian kernel g interpolates the pixels “missing” after up-sampling. The resulting image is subtracted from the Gaussian smoothed image Gk. To rebuild the original image, the process is reversed as Figure “Three-level Gaussian and Laplacian Pyramid” shows.
The Gaussian image pyramid on the left is used to create the Laplacian pyramid in the center, which is used to reconstruct the Gaussian pyramid and original image on the right. In Figure “Three-level Gaussian and Laplacian Pyramid”, I is the original image, Gk is the Gaussian image, Lk is the Laplacian image, subscripts k denote level of the pyramid, g is a Gaussian kernel used to convolve the image before down-sampling or after up-sampling.
The Intel IPP implement only Gaussian pyramids. To build Laplacian pyramids, the following general rule for Intel IPP pyramids functions is to be followed:
Li = Gi - PyrUp(PyrDown(Gi)),
where Li is an image on the i-th level in the Laplacian pyramid, G i is an image on the i-th level in the Gaussian pyramid.