Gaussian Pyramid

by Marco Zennaro

This document describe the implementation of two algorithms. The first one, called "gaussianPyramid" has been implemented with Matlab and it follows the step on page 160 of [Forsyth 2000]:
Start with the original image as the finest layer
For each layer, going from finest to coarsest
    	obtain this layer by smoothing the next finest with a Gaussian
	subsample it
The matlab code for gaussianPyramid can be downloaded from [here].

You can also dowload the implementation of a second algorithm, called "naiveGaussianPyramid" that does the sampling without applying the gaussian filter. It has been implemented in order to be able to have something to compare the gaussianPyramid with. The matlab code for naiveGaussianPyramid can be downloaded from [here].

The two algorithms has been applied on the same picture (that can be downloaded from [here]). The computed pyramids are displayed in figure p1. We used a kernel of size 10 and standard deviation 1 to create the pyramid on the right.

Fig. p1: a gaussian pyramid (on the right) compared with a pyramid obtained by samplying without smoothing

References [Forsyth 2000] D. Fosyth, J. Ponce, "Computer Vision: a modern approach", Prentice Hall, 2000