Author Description

The properties of image scale and its implications for image matching algorithms are introducedin this chapter. This is followed by a brief explanation of the concept of scale-spaceand how both Gaussian and differential scale-spaces can be constructed. Building on theseconcepts, we introduce the multi-resolution image pyramid data structure and how this canbe parameterized in order to build Gaussian and Laplacian of Gaussians image pyramids.These pyramid structures provide for efficient representation and computation withinscale-space and constitute an essential prerequisite to the formation of general purpose imagematching algorithms. The notion of subdividing the scale-space within the levels of thepyramid is also presented, such that any specified degree of continuity (in scale) betweenpyramid levels can be achieved. Examples of image pyramids coded in both the C++ andMatlab programming languages illustrate the practical compromises that must be resolved inpractice when building pyramids.