Chapter 7: The scale-space primal sketch

Chapter 7 in Scale-Space Theory in Computer Vision presents a representation for making explicit image structures in scale-space as well as the relations between image structures at different scales. The representation is based on blobs that are either brighter or darker than the background. At any scale in scale-space grey-level blobs are defined at that scale. Then, these grey-level blobs are linked across scales into objects called scale-space blobs. The relations between these blobs at different scales define a hierarchical data structure called the scale-space primal sketch, and it is proposed that the volume of a scale-space blob in scale-space constitutes a natural measure of blob significance.

To enable comparisons of significance between structures at different scales, it is necessary to measure significance in such a way that structures at different scales are treated in a uniform manner. It is shown how a definition of a transformed scale parameter, effective scale, can be expressed such that it gives intuitive results for both continuous and discrete signals. The volumes of the grey-level blobs must be transformed in a similar manner. That normalization is based on simulation results accumulated for a set of reference signals.

Responsible for this page: Tony Lindeberg