statismoGPmodel: expands a models variability by adding a Gaussian kernel...

In zarquon42b/RvtkStatismo: Integrating statismo and R using the vtkStandardMeshRepresenter

Description

expands a models variability by adding a Gaussian kernel function to the empiric covariance matrix and builds a low-rank approximation of the resulting PCA

Usage

statismoGPmodel(model, kernel = GaussianKernel(50, 10), ncomp = 10,
  nystroem = 500)

Arguments

model	shape model of class pPCA
kernel	an object of class matrixKernel
ncomp	integer: number of PCs to approximate
nystroem	number of samples to compute Nystroem approximation of eigenvectors

Value

returns a shape model of class pPCA

See Also

pPCA, pPCA-class

Examples

### this is a silly example with only 10 landmarks
require(Morpho)
data(boneData)
align <- rigidAlign(boneLM)$rotated
mod <- statismoBuildModel(align)
##extend flexibility using two Gaussian kernels
## Get the empirical kernel first
isoKernel <- IsoKernel(DrawMean(mod),scale=0.001)
kernel1 <- GaussianKernel(10,1)
kernel2 <- GaussianKernel(1,1)
combinedKernel <- SumKernels(kernel1,kernel2)
combinedKernelNoEmp <- SumKernels(combinedKernel,isoKernel)
combinedKernel <- SumKernels(combinedKernelNoEmp,StatisticalModelKernel())
GPmod <- statismoGPmodel(mod,kernel=combinedKernel)
##extend flexibility using two Gaussian kernels but ignoring empiric covariance.
GPmodNoEmp <- statismoGPmodel(mod,kernel=combinedKernelNoEmp)
PC1orig <- DrawSample(mod,2)# get shape in 2sd of first PC of originial model
PC1 <- DrawSample(GPmod,2)# get shape in 2sd of first PC of the extended model
PC1NoEmp <- DrawSample(GPmodNoEmp,2)# get shape in 2sd of first PC
##visualize the differences from the mean (green spheres)
## Not run: 
deformGrid3d(PC1,DrawMean(GPmod),ngrid=0)##
##only deviates in 5 landmarks from the mean (dark blue)
deformGrid3d(PC1NoEmp,DrawMean(GPmod),ngrid=0,col1=4,add=TRUE)
deformGrid3d(PC1orig,DrawMean(GPmod),ngrid=0,col1=5,add=TRUE)

## End(Not run)

zarquon42b/RvtkStatismo documentation built on May 4, 2019, 9:09 p.m.