Title :
Geometric moments in scale-spaces
Author :
Bresson, Xavier ; Vandergheynst, Pierre ; Thiran, Jean-Philippe
Author_Institution :
Signal Process. Lab., Swiss Fed. Inst. of Technol., Lausanne, Switzerland
Abstract :
We present a generalization of geometric moments in scale-spaces derived from the general heat diffusion equation, with a particular interest in minimax flow. As an application of those theoretical developments, two multiscale moments are used to enhance the classical Euclidean registration process. They are computed from a multiscale representation which preserves the global shape of the objects, clearly outperforming the classical Euclidean moment-based object registration.
Keywords :
geometry; image registration; image representation; partial differential equations; classical Euclidean registration process; geometric moments; global shape preservation; min/max flow; scale-spaces; Convolution; Diffusion processes; Image analysis; Kernel; Laplace equations; Linearity; Performance analysis; Shape; Signal processing; Smoothing methods;
Conference_Titel :
Pattern Recognition, 2002. Proceedings. 16th International Conference on
Print_ISBN :
0-7695-1695-X
DOI :
10.1109/ICPR.2002.1048328