Commentary to “Morphological Image Analysis and Its Application to

Commentary to
Morphological Image Analysis and Its
Application to Sunspot Classification
Ricardo Vilalta
Dept. of Computer Science
University of Houston
June, 2011
Subjective Nature of Classification
Representation is crucial to produce a well-defined classification problem.
Mount Wilson scheme seems too subjective and needs a more clear and detailed
class description .
Using a single classification technique is risky; given the no. of available
classifiers, it is a better practice to use at least 5-6 different techniques (decision
forests can fall into over-fitting).
Bias Variance and Irreducible Error
In general, the expected prediction error EPE:
EPE (xo) = Var(y0 | x0) + VarT (y0) + Bias2 (y0)
Bias Variance and Irreducible Error
In general, the expected prediction error EPE:
EPE (xo) = Var(y0 | x0) + VarT (y0) + Bias2 (y0)
High variance due
to rich family of functions; unstable
Bias Variance and Irreducible Error
In general, the expected prediction error EPE:
EPE (xo) = Var(y0 | x0) + VarT (y0) + Bias2 (y0)
High bias due to poor
family of functions
but very stable
High Bias Low Variance
Low Bias High Variance
A Trade-Off Between Bias and Variance
Bias Variance and Irreducible Error
In general, the expected prediction error EPE:
EPE (xo) = Var(y0 | x0) + VarT (y0) + Bias2 (y0)
P(y1|x)
P(y2|x)
Bayes Error
x
The Importance of Contextual
and Spatial Information
Application on Mars
Collaborator:
Lunar and
Planetary Institute
Planetary Scientist: Tomasz Stepinski
Objective: automated creation of
geomorphic maps
Martian landscape
Geomorphic map
shows landforms
chosen and
defined by a
domain expert.
Digital Elevation Map
Manually drawn geomorphic map of this landscape
Geomorphic Map
Perspective View.
Segmentation: Results.
2631 segments
homogeneous in
slope, curvature
and flood.
Displayed on an elevation background.
Segmentation: Results.
Classification: Labeling.
A representative subset of objects
are labeled as one of the following
six classes:
¤  Plain
¤  Crater Floor
¤  Convex Crater Walls
¤  Concave Crater Walls
¤  Convex Ridges
¤  Concave Ridges
Labeled segments.
Classification: Ground Truth
Classification: Support Vector Machines
Classification: Refined Vector Machines
The Interdisciplinary Nature of Astronomy
High Performance
Computing
Probability &
Statistics
Databases Systems
Astronomy
Machine
Learning
Artificial
Intelligence
Image
Processing
THANK YOU