06 - Shape Analysis

Question 1

Component labeling

Answer 1

• segmentation

- label = uniquely id obj = segmented pixels touching each other (using neighborhood)

Question 2

Metrics

Answer 2

• center of volume
• center of mass (intensity)
• area, perimeter, etc.
• regionprops

Question 3

Statistical tools

Answer 3

• PCA = find a linear combination that best describes the variations in the system
• covariance matrix, eigentransform, eigenvectors and eigenvalues
• PC = obj orientation, scores = corresponding magnitude
• elliptical model = rpstation, eigenvectors = semiaxes with length prop to sqrt(eigenvalue)

Question 4

Meshing

Answer 4

• process of turning a connected set of pixels into a list of vertices and edges
• need connectivity info
• marching cubes: add faces one voxel at a time to incorporate the most simple surface which would explain the values

Question 5

Distance maps

Answer 5

• metrics: euclidian, weighted metrics
• 4-connected = cityblock/Manhattan; 8-connected = checkerboard
• /!\ to anisotropic voxels, use custom distance map or convert to isotropic during filtering
• fg to bg = info obj size/shape; bg to fg = info about dist b/ obj
• relatively insensitive to small changes in connectivity

Question 6

Thickness maps

Answer 6

• assessing size and structure of obj
• for each voxel finds the largest sphere that contains it and is included in the obj
• for pore radius mapping

Question 7

Curvatures

Answer 7

• K = 1/R
• in 3D: 2 principal curvatures
• gives characteristic shape

Question 8

Texture analysis

Answer 8

• texture = partly periodic, partly stochastic
• characterize = co-occurence matrix
• tiling = divide img to indvdl tiles for further processing
  metrics: dissimilarity, correlation; best correlation with what we’re trying to segment?