Effective classification of 3D closed surfaces: Application to modeling neuroanatomical structures

We present a new technique for 3D surface object classification that combines a powerful shape representation approach with suitable pattern classification techniques. Spherical harmonic parameterization and normalization techniques are used to describe a surface shape and derive a dual high dimensional landmark representation. A point distribution model is applied to reduce the dimensionality. Linear discriminants and support vector machines are used for classification, with several feature selection approaches. We demonstrate the effectiveness of this technique using a hippocampus dataset and achieve a jackknife classification accuracy of 93%.