Realtime scale selection in hybrid multiscale representationsTony Lindeberg and Lars BretznerTechnical report CVAP279, ISRN KTH NA/P03/07SE. Department of Numerical Analysis and Computer Science, KTH (Royal Institute of Technology), SE100 44 Stockholm, Sweden, June 2003.Shortened version in Proc. ScaleSpace'03, Isle of Skye, Scotland, Springer Lecture Notes in Computer Science, volume 2695, pages 148163 AbstractLocal scale information extracted from visual data in a bottomup manner constitutes an important cue for a large number of visual tasks. This article presents a framework for how the computation of such scale descriptors can be performed in real time on a standard computer.The proposed scale selection framework is expressed within a novel type of multiscale representation, referred to as hybrid multiscale representation, which aims at integrating and providing variable tradeoffs between the relative advantages of pyramids and scalespace representation, in terms of computational efficiency and computational accuracy. Starting from binomial scalespace kernels of different widths, we describe a family pyramid representations, in which the regular pyramid concept and the regular scalespace representation constitute limiting cases. In particular, the steepness of the pyramid as well as the sampling density in the scale direction can be varied. It is shown how the definition of $\gamma$normalized derivative operators underlying the automatic scale selection mechanism can be transferred from a regular scalespace to a hybrid pyramid, and two alternative definitions are studied in detail, referred to as variance normalization and $l_p$normalization. The computational accuracy of these two schemes is evaluated, and it is shown how the choice of subsampling rate provides a tradeoff between the computational efficiency and the accuracy of the scale descriptors. Experimental evaluations are presented for both synthetic and real data. In a simplified form, this scale selection mechanism has been running for two years, in a realtime computer vision system. PDF: (TR 362 kb) (Shortened 220 kb) PostScript: (TR 482 kb) (Shortened 393 kb)
Related publications: (General scale selection principle) (Discrete scalespace theory on a spatial domain based on noncreation of local extrema) (Discrete scalespace theory on a spatial domain based on nonenhancement of local extrema) (Review paper on principles for automatic scale selection) (Monograph on scalespace theory)
Responsible for this page:
Tony Lindeberg
