Kernel PCA for similarity invariant shape recognition
Neurocomputing
Wavelet, active basis, and shape script: a tour in the sparse land
Proceedings of the international conference on Multimedia information retrieval
Editor's choice article: On growth and formlets: Sparse multi-scale coding of planar shape
Image and Vision Computing
| Hi-index | 0.00 |
This paper presents a multi-scale generative model for representinganimate shapes and extracting meaningful parts of objects. Themodel assumes that animate shapes (2D simple closed curves) areformed by a linear superposition of a number of shape bases. Theseshapebases resemble the multi-scale Gabor bases in image pyramidrepresentation, are well localized in both spatial and frequencydomains, and form an over-complete dictionary. This model issimpler than the popular B-spline representation since it does notengage a domainpartition. Thus it eliminates the interferencebetween adjacent B-spline bases, and becomes a true linear additivemodel. We pursue the bases by reconstructing the shape in acoarse-to-fine procedure through curve evolution. These shape basesare further organized ina tree-structure where the bases in eachsubtree sum up to an intuitive part of the object. To buildprobabilistic model for a class of objects, we propose a Markovrandom field model at each level of the tree representation toaccount for the spatial relationship between bases. Thus the finalmodel integrates a Markov tree (generative) model over scales and aMarkov random field over space. We adopt EM-type algorithm forlearning the meaningful parts for a shape class, and show someresults on shape synthesis.