Chapter 10: Extracting salient image structures

Chapter 10 in Scale-Space Theory in Computer Vision experimentally demonstrates how the scale-space primal sketch can be used for extracting significant blob-like structures from image data as well as associated scale levels for treating those. By sorting the basic primitives in the scale-space primal sketch, the scale-space blobs, with respect to their normalized volumes in scale-space, a ranking on significance is obtained. Such descriptors constitute coarse segmentation cues, and can serve as regions of interest to other processes.

The treatment is based on two basic assumptions;

• in the absence of other evidence, structures, which are significant in scale-space, are likely to correspond to salient structures in the image, and
• in the absence of other evidence, scale levels can be selected where the blob response assumes its maximum over scales.

Experimental results are presented for different types of real imagery demonstrating that the methodology gives intuitvely reasonable results.