Chapter 11: Guiding visual processes with scale-space

Chapter 11 in Scale-Space Theory in Computer Vision demonstrated how the qualitative scale and region descriptors extracted by the scale-space primal sketch can be used for guiding other processes in early vision and for simplifying their tasks.

An integration experiment with edge detection is presented, where edges are detected at coarse scales given by scale-space blobs, and then tracked to finer scales in order to improve the localization. In histogram analysis, the scale-space primal sketch is used for automatic peak detection. More generally, such descriptors can be used for guiding the focus-of-attention of active vision systems. With respect to a test problem of detecting and classifying junctions, it is demonstrated how the blobs can be used for generating regions of interest, and for providing coarse context information (window sizes) for analysing those.

Finally, it is briefly outlined how the scale-space primal sketch can be applied to other visual tasks such as texture analysis, perceptual grouping and matching problems. Experiments on real imagery demonstrate that the proposed theory gives intuitively reasonable results.

Responsible for this page: Tony Lindeberg