Lecture 7 Flashcards
Image registration
The changing of images such that both images are part of the same coordinative systems
- Aligning for colocalization
- Fixating cells
- Mosaicking
- Multimodal fusion
- Longitudianal analysis
Aligning for colocalization
Can be used for chomatic aberration. Less double and more clear
Fixating cells
Makes it easier to study their internal changes like shape and position. You can segment them out and map them on a grid
Mosaicking
Enlarge the field of view. This way you can see a large picture. You need them to overlap
Multimodal fusion
This can help with joint analysis. For example fluorescence and electron microscopy mixing
Longtitudinal analysis
Can compare different time points. Need to be stacked in the same positional state
Extrinsic vs intrinsic information
Extrinsic if the images are very different from one another. Need external markers to allign the images. Otherwise intrinsic information can be used. Look at the Pearson’s correlation coefficient
landmarks
intensity differences
Can be improved by extracting common image features such as edges or ridges
Joint histogram
To look at intesity differenes
Similarity measures
- Normalized cross-correlation
- Sum of squared differences
- Mutual information
Normalized cross-correlation
Sum of squared differences
Mutual information
Geometric image transformations
- Rigid transformations
- Affine transformations
- grid based nonrigid transformations
Rigid transformations
- Translation
- Rotation
Affine transformations
Perserves points, straight lines and parallelism between straight lines.
- Shearing
- Scaling
- Identity
- Translation
- Rotation
Grid based non rigid transformations
- Random
- Elastic
Define a coarse grid of control points and estimate the local displacements. Smoothly interpolate estimated displacements to all coordinates and apply.
Optimization
- Iterative
- Brute force
- Gradient based
- Quadratic
- Multiscale optimization
Iterative optimization
Step size is important, determines the accuracy.
Brute force
Simply try all possible parameter values within certain limits.
Gradient based
Move toward the optimum of a local quadratic approcimation of M
Quadratic optimization
Move toward the optimum of a local quadratic approximation of M
Multiscale optimization
Iterative registration from low to original imare resolution. More efficient, but more robust.
Interpolation methods
- Nearest-Neighbor
- Linear
- Cubic interpolation
Atlas based registration
Minimize overlap error of full skeleton, then do this for all bones seperately. Now normalize the poster to comper the structure to other posters
