Title: IPAM-UCLA Tutorial May 14-18, 2001 GBM in Image Processing, Computer Vision, and Computer Graphics
1IPAM-UCLA TutorialMay 14-18, 2001 GBM in Image
Processing,Computer Vision, andComputer
Graphics
- Guillermo Sapiro
- Electrical and Computer Engineering
- University of Minnesota
- guille_at_ece.umn.edu
Supported by IPAM, NSF, ONR
2GBM Moving Curves
3Basic curve evolution
- Planar curve
- General flow
- General geometric flow
4Mathematical morphology
- Classical theory, based on Minkowsky addition.
- The old and (probably wrong) way of doing
geometric image analysis. - Has very important lessons to learn!!!!
- Basic definitions
- A Image in Euclidean space (R or Z)
- B Structuring element (symmetric)
- Nothing else than Minkowsky addition
5Mathematical morphology Definitions
6Mathematical morphology Is it good or bad?
- Advantages
- Nice mathematical properties (set theory)
- Extension to Lattices
- Disadvantages
- Discrete Minkowsky addition does not look good,
has to be replaced by better ways of computing
discrete distances. - Major important concept Level-sets
- Commutes with thresholding (level-sets) Do
binary on each level sets or do gray-level on all
the image gt same result - It is in certain sense a particular case of curve
evolution (before the lattices part)
7Mathematical morphology via curve evolution
Huygens principle
8Mathematical morphology via curve evolution
(cont.)
- General velocity
- Examples
- Nothing else than changing the metric (distance).
- Can be explained also based on dynamic
programming and time of arrival - See Sapiro et al., Brocket-Maragos, Alvarez et
al., Evans, Falcone
9Planar differential geometry
- Euclidean invariant parametrization
-
- Affine invariant parametrization
C
s
(
)
1
s
C
C
,
lt
gt
0
s
ss
C
C
s
ss
k
C
ss
10Planar differential geometry (cont.)
- Curvature constant for circles or straight lines
(0) - Curvature defines curve up to Euclidean motion
- At least 4 points with dk/ds0
- Defined for all curves
- Curvature constant for ellipses (gt0), hyperbolas
(lt0), and parabolas (0) - Curvature defines curve up to affine motion
- At least 6 points with dk/ds0
- Defined only for convex curves segment at
inflection points
11Planar differential geometry (cont.)
X
C(s)
Distance has a local extrema iff X is on the
normal
123D Differential geometry
- Remember mean and Gaussian curvatures?
- Each regular surface has two principal
curvatures. The average is the mean curvature,
the product the Gaussian. These are also related
to the tangential map, etc, etc. See DoCarmo for
details.
13Riemannian geometry, Lie theory
- What about other non-Euclidean metrics?
- What about invariants to other (Lie) groups,
e.g., projective? - What about differential invariants?
Semi-differential invariants? Are there any
general theories?
14Smoothing by classical heat flow
- Linear
- Equivalent to Gaussian filtering
- Unique linear scale-space
- Non geometric
- Shrinks the shape
- Implementation problems
15Invariant shape deformations
- Formulate shape deformations
- Geometric
- Invariant to camera transform
- The best possible
- Change only the desired features
- Motivation
- Mathematics
- From static differential geometry to dynamic
- Beautiful
- Computer vision and image processing
- Invariant shape segmentation and analysis
- Image processing via image deformations
- Robotics
- Motion planning
- Accurate geometric object detection and tracking
- Robot manipulation and grasping
16Basic planar differential geometry
- For every Lie group we will consider, exists and
invariant parametrization s, the group arc-length - For every such a group exists and invariant
signature, the group curvature, k
Low curvature
High curvature
Negative curvature
17What and why invariant
- Camera
- motion
Deformation - Camera/object movement in the space
- Transformations description (for flat objects)
- Euclidean
- Motion parallel to the camera and planar
projection - Affine
- Planar projection
- Projective
18Euclidean geometric heat flow
- Use the Euclidean arc-length
- The deformation
- Smoothly deforms to a circle (Gage-Hamilton,
Grayson) - Geometric smoothing
- Reduces length as fast as possible
-
19Affine geometric heat flow (Sapiro-Tannenbaum)
- Use the affine arc-length
- The flow
20Affine geometric heat flow (cont.)
- Geometric smoothing (preserving area if desired)
- Total curvature decreases
- Maxima of curvature decreases
- Number of inflections decreases
- Smoothly deforms a shape into an ellipse
- Decreases area as fast as possible (in an affine
form) - Existence also for non-smooth curves
- Viscosity framework (Alvarez-Guichard-Morel-Lions)
- Polygons (Angenent-Sapiro-Tannenbaum)
- Applications
- Curvature computation for shape recognition
reduce noise (Morel et al.) - Simplify curvature computation (Faugeras 95)
- Object recognition for robot manipulation
(Cipolla 95)
21General invariant flows
- Theorem For every sub-group of the projective
group the most general invariant curve
deformation has the form - Theorem In general dimensions, the most general
invariant flow is given by - u graph locally representing the surface
- g invariant metric
- E(g) variational derivative of g
- See Olver et al., Alvarez et al., Caselles-Sbert
22General Geometric Framework For Object
Segmentation
23Introduction
- Goal Object detection
- Approach Curve/surface deformation
- Geometry dependent regularization
- Image dependent velocity
- Characteristics
- Unifies previously considered independent
approaches - Relates segmentation with anisotropic diffusion
- General
- Any topology
- Any type of image data
- Any dimension
- Holds formal results
24Notation
25Basic active contours approach
- Terzopoulos et al., Cohen et al.
- Drawbacks
- Too many parameters
- Non-geometric
- Handling topology changes
26Geodesic active contours (Caselles-Kimmel-Sapiro)
- Generalize image dependent energy
- Eliminate high order smoothness term
- Equal internal and external energies
- Maupertuis and Fermat principles of dynamical
systems
27Geodesic computation
- Gradient-descent
- Level-sets (Osher-Sethian)
28Further geometric interpretation
29Model correctness
- Theorem The deformation is independent of the
level-sets embedding function - Theorem There is a unique solution to the flow
in the viscosity framework - Theorem The curve converges to ideal objects
when present in the image - Related work
- Kimia-Tannenbaum-Zucker
- Caselles et al.
- Malladi-Sethian-Vemuri
- Kichenassamy at al.
- Tek-Kimia, Whitacker
- New work
- Chan-Vese
- Paragios-Deriche
- Yezzi et al.
- Faugeras et al.
30Extensions
- Gray-level values
- ds - length element (geodesics)
- Ordinary edge detector (gradient)
- Surfaces
- ds - area element (minimal surfaces)
- 3D edge detector
- Vector-valued images (color, texture, medical,
etc) - ds - length element
- Vector-valued edge detector (vector geodesics)
- Eigenvalues of the first fundamental form in
Riemannian space - Invariant detection (affine area geodesics)
- ds - affine length element (area related)
- Affine invariant edge detector
- Affine norm for gradient descent
31Why color edges?
32Notation
- Image
- Texture Gabor decomposition
-
33Color edge computation
- Given a metric (Euclidean)
- Compute first fundamental form
- Compute eigenvectors and eigenvalues
- Edge maximal eigenvalue and its eigenvector
- Basic properties
- Eigenvectors are orthonormal
- Minimal eigenvalue is not zero
34GBM Moving Images
35Anisotropic diffusion
Isotropic vs. Anisotropic Smoothing
36Isotropic diffusion (Koenderink, Witkin)
- All equivalent
- Gaussian filtering of the image
- Heat flow
- Minimize the L2 norm
37Isotropic diffusion Good things
- Gaussian filtering if and only if
- Linear
- Shift-invariant
- No creation of zero crossings
- Gaussian filtering if and only if
- Linear
- Shift-invariant
- Semi-group property
- Scale-invariant (dimensionless)
- Unique linear filter that defines a scale-space
Do not creates information at coarser scales - Where everything started (Koenderink, Witkin)
38Isotropic diffusion Bad things and possible
solutions
- Non-geometric
- Problems with implementations
- Who said linear? Replace heat flow by parabolic
PDEs (Hummels original idea) - Why parabolic? Because of the maximum principle.
39Perona-Malik anisotropic diffusion
- Replace the L2 by a different norm (e.g., L1,
Rudin-Osher-Fatemi Lorentzian, Black et. al.
etc)
40Selection of stopping term h
- How do we select h?
- hxx gt L2 gt linear gt Isotropic
diffusion - hx gt L1 (Rudin-Osher-Fatemi)
41Robust anisotropic diffusion
- General theory for selection h, based on the
theory of influence functions in robust
statistics - Edges should be considered outliers At certain
point, h, the influence, should be zero.
42Directional diffusion
- Diffuse in the direction perpendicular to the
edges (Avarez et al.)
43From active contours to anisotropic diffusion
- Replace embedding function in level-sets
formulation by image itself
Shock-filters (Osher-Rudin)
Anisotropic diffusion (Alvarez et al.)
44Relation with Perona-Malik anisotropic diffusion
Total variation, Robust estimation
Anisotropic diffusion
45Concluding remarks
Terzopoulos snakes
Geometric interpretation Dynamical
systems Level-sets
Terms elimination
Curve evolution active contours
Geodesic active contours
Use image as embedding
Geometric diffusion
Self-snakes
Mumford-Shah
Add
Shock-filters
Divide by gradient
Perona-Malik flow
Variational interpretation
Total Variation
Robust Estimation
46Anisotropic Diffusion of the Posterior
47ADP in MRI
Review MAP Estimation
- 3 classes sulcus, gray matter, white matter
- Prior probability Pr(classC)
- Posterior probability Pr(classC data)
- MAP Choose class C that maximizes posterior
- C arg max Pr(classC data)
- C
- Bayes Rule
- Pr(classC data) Pr(data
classC).Pr(classC) - Pr(data)
- What is our prior, Pr(classC)?
48ADP Common Techniques
MAP Estimation Uniform Prior
Classification
49ADP Results
Anisotropic smoothing of posterior
(Teo-Sapiro-Wandell)
Smoothed Posterior
Classification
Posterior
50ADP Comparisons
Comparison with manual segmentation
Automatic (2 min)
MR Image
Manual (18 hours)
51SAR segmentation via vector probability
processing
With A. Pardo (see also Haker-Sapiro-Tannenbaum)
52Anisotropic Diffusion in Vector Space
53Goal and approach (Ringach-Sapiro)
- Goal
- Enhancement of vector valued data
- Extend classical theories of scalar PDEs in
image processing - Approach
- Work in vector space
- Compute vector edges
- Anisotropic diffusion
- Important Works for any vector data
- See also Cumani, Di Zenzo, Chambolle
54Notation
- Image
- Texture Gabor decomposition
-
55Color edge computation
- Given a metric (Euclidean)
- Compute first fundamental form
- Compute eigenvectors and eigenvalues
- Edge maximal eigenvalue and its eigenvector
- Basic properties
- Eigenvectors are orthonormal
- Minimal eigenvalue is not zero
56Color anisotropic diffusion
- Direction Minimal change
- Strength
57Level lines for vectorial images (Chung-Sapiro)
Vector and scalar representation sharing
level-lines
58Contrast Enhancement (Sapiro-Caselles, and
Caselles-Lisani-Morel-Sapiro)
- Contrast enhancement via image deformations
- Approach Histogram modification
- Characteristics
- Simultaneous contrast enhancement and denoising
- First explanation of histogram modification in
image domain - Extended to local
- First semi-global partial differential equation
in image processing - Formal existence results
59GBM Beyond the flat manifolds
60The main problem and our goal (Tang-Sapiro-Caselle
s)
- Goal Enhancement and analysis of directional
data (and data on non-flat manifolds) - Problem Directions are unit vectors
- Regular images vs Directions
- Applications
- Optical flow, Gradients
- Vector data (normalized)
- Color image enhancement
- Surface normals and principal directions
- Flows in general manifolds
61Average
62Most popular previous approaches
- Work with angles Operations on the sphere
- Average, median, etc
- Statistics of directional data, Mardia
- Orientation Diffusion, Perona (1998)
- Tensor diffusion
- Weickert, Granlund-Knuttson
- See also Chan-Shen
63Anisotropic Diffusion
Isotropic (Heat equation)
Anisotropic
64What have we learned from images?
Robust Estimation
Robust function
Gradient Descent
Influence function (defines outliers)
Anisotropic Diffusion
65Anisotropic Diffusion
Isotropic (Heat equation)
Anisotropic
66Back to Directions Basic Idea
- Use the theory of harmonic maps
- Find a map I from two manifolds (M,g) and (N,h)
such that - In particular, liquid crystals
67The Gradient-Descent Equations
General (p2)
Liquid crystals
68A Few Theoretical Results (over hundreds relevant)
- For 2D unit vectors (n1), and p2, a unique
solution exists and singularities are isolated
points (if they exists at all). For smooth data,
singularities do not occur. - Singularities occur for 3D unit vectors (p2).
- Singularities well characterized for 1ltplt2.
- Energy well characterized for 1ltplt2.
- No singularities for manifolds with non-positive
curvature.
69Intermezzo Visualizing Directions
- Arrows
- Color Map
- Line Integral Convolution
70Examples (Isotropic)
713D vector (Isotropic)
72Denoising
73(No Transcript)
74(No Transcript)
75Optical flow
76Optical flow (cont.)
77Gradient
78Gradient (cont.)
79Color Image Enhancement
80Color image enhancement (cont.)
81Color image enhancement (cont.)
82Vector probability diffusion (with Alvaro Pardo)
- Perform diffusion on the hyperplane representing
probabilities
83Vector probability diffusion (cont.)
- The numerical implementation also stays on the
hyperplane - The numerical implementation also holds a maximum
and minimum principle
84Vector probability diffusion (cont.)
- Diffuse posterior probabilities (following
Teo-Sapiro-Wandell and Haker-Sapiro-Tannenbaum)
85Current Results and Future Research
- Novel framework for analysis of directional data
- Isotropic and anisotropic
- Works in any dimension
- Supported by theoretical results on existence,
uniqueness, singularity classification - See also Sochen-Kimmel-Malladi IEEE-IP 98,
Chan-Shen UCLA 99, Osher-Vese UCLA 00.