Spherical Self-Organizing Maps

The thesis documents various implementations of the
spherical self-organizing feature map spanning from
range imagery to multi-dimensional satellite imagery.
Other implementations include archaeological data
and face modeling. The techniques are adapted
versions of the fundamental learning strategy of the
self-organizing map, introduced by Dr. Teuvo Kohonen,
and implemented within a spherical topology.
Continuity in learning is imposed by the predefined
spherical lattice which also transforms the data into
a 3D tessellated form. Since the tessellated forms
originate from a sphere they simplify the computation
of transformation parameters, object re-orientation
etc. commonly used in three-dimensional graphics.

Autorentext

Dr. Sangole is a Researcher in Mechanical Engineering at EcolePolytechnique and the Marie-Enfant Rehabilitation Center ofCHU-Sainte-Justine, Montreal, QC. Her research is inmusculoskeletal biomechanics. She is a manufacturing engineerwith previous research experience in deformable modeling, datavisualization and Computer-Aided-Design exchange.

Klappentext

The thesis documents various implementations of thespherical self-organizing feature map spanning fromrange imagery to multi-dimensional satellite imagery. Other implementations include archaeological dataand face modeling. The techniques are adaptedversions of the fundamental learning strategy of theself-organizing map, introduced by Dr. Teuvo Kohonen,and implemented within a spherical topology.Continuity in learning is imposed by the predefinedspherical lattice which also transforms the data intoa 3D tessellated form. Since the tessellated formsoriginate from a sphere they simplify the computationof transformation parameters, object re-orientationetc. commonly used in three-dimensional graphics.