image process.ppt

(1)

(2)

Natural Evolution

Requirements for evolution:

 Representation

 Selection Pressure

 Recombination

(3)

Natural Evolution

 Representation: DNA

 Selection Pressure: Survival,

Reproduction (determined implicitly by environment)

 Recombination: Mating, Replicating

(4)

Natural Evolution:

Panthera leo

 Representation:

– _{Linear DNA} – Diploid cells

 Selection Pressure:

– Rate of Reproduction

 Recombination:

– _{Sexual reproduction}

 Diversity

– _{Prides, with some}

(5)

Natural Evolution:

Streptococcus pyogenes

 Representation:

– _{Circular DNA} – Monoploid cells

 Selection Pressure:

– Rate of Reproduction

 Recombination:

– _{Binary fission (asexual)}

 Diversity

– _{Very little}

(6)

Artificial Evolution: Genetic Algorithms

 Genetic algorithms are arguably the simplest

class of evolutionary algorithms

 GAs make use of simple representation,

reproduction, and diversity mechanisms

 These aspects are usually independent of

(7)

Artificial Evolution: Genetic Algorithms

 Representation: Linear chromosome

(array of values)

 Selection Pressure: Explicit fitness

function,

Selection algorithms

 Recombination: Crossover, Mutation

(8)

(array of values)

0 1 1 0 1 1 0 1 0 1 0 1

(9)

(array of values)

(10)

function

Fitness function is problem-specific. The only limits are those of imagination and

practicality.

(11)

Artificial Evolution: Genetic Algorithms

 Selection Pressure: Selection algorithms

Selection algorithms are usually applied to choose which individuals reproduce…

(12)

Common selection algorithms include: Roulette wheel selection,

Rank-based selection,

(13)

 Recombination: Crossover

Parent 0

Parent 1

(14)

Artificial Evolution: Genetic Algorithms

Parent 0

Parent 1

Child 0

Child 1

(15)

Parent 0

Parent 1

Child 0

Child 1

(16)

Parent 0

Parent 1

Child 0

Child 1

(17)

 Recombination: Mutation

Parent 0

Point Mutation

(18)

Artificial Evolution: Genetic Algorithms

 Recombination: Duplication

Parent 0

Used for Elitism

(19)

 Diversity: Fixed-size population

Generational GAs use individuals from current generation to create an entirely new

generation of the same size.

(20)

GA Example: Noise Reduction

(21)

GA Example: Noise Reduction

 Reasonable choices:

– Pick a window size of n×n

– Use a real-valued chromosome of length n2

– Use a blending crossover

– Use a modest mutation operator

– Apply mask to noisy images

– Set fitness to the Euclidean distance between

(22)

 Would we expect the evolved solution to

work well for B&W noise?

 Would we expect the evolved solution to

(23)

GA Example: Noise Reduction

 “You get what you evolve for!”

 So write your fitness function carefully.

(24)

 Barring some much more creative way of

using the linear chromosome, we can’t evolve a simple median filter using this representation.

 Wouldn’t it be nice if we could evolve

(25)

Artificial Evolution: Genetic Programs

 Genetic programs are a (slightly) more

complex class of evolutionary algorithms.

 The main difference is in the representation,

(26)

Artificial Evolution: Genetic Programs

 Representation: Program tree

(27)

The user must specify a function set and a terminal set.

1

^ if ~ x y z

v | 0

(28)

Artificial Evolution: Genetic Programs

Chromosome can be Boolean-valued, integer-valued, real-integer-valued, complex-integer-valued, etc.

(29)

Chromosome can also be strongly-typed:

(30)

function

– Problem specific, same as GA

(31)

Artificial Evolution: Genetic Programs

(32)

(33)

(34)

 Recombination: Mutation

Parent 0 Child 0

Grow an entire

(35)

 Diversity: Population

(36)

GP Example: Landslide Detection

 From Rosin and Hervás’ 2002 paper

Image Thresholding for Landslide Detection by Genetic

(37)

GP Example: Landslide Detection

 Focused on the task of identifying landslide

areas in an image using “before” and “after” satellite images from Veneto, Italy.

 Desired output is a binary image with the

pixels in the landslide area colored black and all other pixels white.

 Previous work using more conventional

(38)

GP Example: Landslide Detection

(39)

 Representation: Real-valued program tree

 Function set: *, +, -, /, abs, sigmoid, min, max

 Terminal set: random constants, difference image pixel values, smoothed difference

values, pixel values from various other transforms

(40)

 Interpretation: Positive = landslide

Negative = no landslide 0? (not reported in paper)

 Fitness function: % of correctly classified pixels  Selection algorithms:

Not reported in paper

 Recombination:

(41)

GP Example: Landslide Detection

 Diversity: Generational GP

 Population size: 20,000

(42)

 Results: best-of-run

 s10 is Gaussian

smoothing with st. dev. of 10

 o15 is “opened” image

with a 15x15 mask

 dt is distance

transformed pixel value

 difference is the

(43)

Images from Rosin and Hervás (2002).

(44)

GP Example: Landslide Detection

 Problems with the methodology:

– Very few details, would be difficult to reproduce

– Only used a portion (3%) of a single image for

training

– Used the same image for evaluation!!!

– Did not report fitness scores for experiments, rely

(45)

GP Example: Landslide Detection

 Strong points of the research:

– Shows benefit of using pre-processed inputs

– Evolved “better” classifier than the authors were

able to design themselves

– Shows that intuitive understanding of final result is

(46)

Concluding points

 Evolutionary algorithms such as GP may be

suitable for evolving, rather than designing, image processing algorithms

 Evolutionary algorithms are not guaranteed

to produce an optimal solution

 Evolutionary algorithms may take a long time

(47)

(48)

Resources

 Dawkins, Richard. 1976. The Selfish Gene. Oxford University

Press, Oxford, UK.

– _{Introduces Dawkin’s selfish gene theory, which argues that the}

gene- not the individual or the species- is the unit of evolutionary selection. Extremely important for understanding natural evolution, with some ramifications for artificial evolution, as well.

 Eiben, A.E. and J.E. Smith, ed. 1998. Introduction to

Evolutionary Computing. Springer-Verlag, Berlin, Germany.

– _{Provides very brief introductions to all of the major classes of}

(49)

Resources

 Goldberg, D.E. 1989. Genetic Algorithms in Search,

Optimization, and Machine Learning. Addison-Wesley.

– _{The “bible” of the simple GA. Focuses on the Boolean (bit string)}

representation and gives theoretic justifications for its success.

 Holland, J.H. 1975. Adaptation in Natural and Artificial

Systems. The University of Michigan Press, Ann Arbor, MI.

– _{From the man who is credited with inventing the genetic algorithm}

and birthing the field of evolutionary algorithms. (Though I have read that von Neumann suggested the idea of evolutionary

(50)

Resources

 Koza, J.R. 1992. Genetic Programming: On the Programming

of Computers by Means of Natural Selection. MIT Press, Cambridge, MA.

– _{Introduces the genetic program. Some earlier work can arguably}

be said to fall under the title “genetic program,” but Koza was the first to treat it as a rigorous methodology. The work is a tour de force in the technique, using GPs to efficiently solve a broad range of problems. 800+ pages, but it reads very quickly.

 _{Koza, J.R. 1992. Genetic Programming II: Automatic Discovery}

of Reusable Programs. MIT Press, Cambridge, MA.

– _{Introduces the}_{automatically defined function}_{. ADFs allow for the}

(51)

Resources

 Rosin, P.L. and J. Hervás. 2002. Image Thresholding for