genetic algorithms and image understanding sam clanton computer integrated surgery ii march 14, 2001

Post on 18-Jan-2016

218 Views

Category:

Documents

0 Downloads

Preview:

Click to see full reader

TRANSCRIPT

Genetic Algorithmsand Image Understanding

Sam Clanton

Computer Integrated Surgery II

March 14, 2001

Resources

• Bhanu, Bir and Lee, Sunkee. Genetic Learning for Adaptive Image Segmentation. Kluwer Academic Publishers, 1994

• Goldberg, David. Genetic Algorithms in Search, Optimization, and Machine Learning. Addison Wesley Longman, 1989.

Genetic Algorithms

• Optimization Problems

• Adaptive Systems

• Speed-Critical Applications

Are Useful For…

General Problem to be Solved

• The k-armed bandit problem

Picture: Goldberg

How do we maximize our winnings?

GA’s are good for multiple, many-armed bandits.

What Is a Genetic Algorithm?

• Operates on principle of

survival of the fittest

• “Population Pool” of Parameters

• Genetic Operators - Reproduction, Crossover, and Mutation

Survival Of the Fittest

• Analogous to survival in biological system

• Fitness Function

• Optimization == Finding most fit parameter set for a particular problem

Selk(an elk) ~ Ability to run away (elk, lions, tigers)Ability to run away (herd, lions, tigers)

Spset(a pset) ~ Ability to perform task(pset, input)Ability to perform task(population, input)

Population Pool

24, 32, 76, 1

34, 43, 6, 17

• “Surviving” parameter sets are kept around

• Individuals are extracted and applied when input resembles past input for that individual.

• Genetic operators add new individuals to pool

• Individuals can be dropped when they appear useless

Genetic Operators

• Affect survival of particular schema

Schema - string representation of a feature

• Reproduction f(H) / favg

• Crossover 1 - pc * L(H) / L(total)• Mutation 1 - L(H) * pm

Feature Preservation

• Overall Equation

m(H, t+1) = m(H, t) * F(H)/favg

Reproduction

* (1 - pc L(H) / L(tot)

Crossover

- L(H) * pm )

Mutation

An Example - Reproduction

String Initial Pop

X val F(x) = x * x

Pselect (fitness/ total fitness)

Exp. Count

(fitness / avg fitness)

Actual Count (roulette)

1 01101 13 169 .14 .58 1

2 11000 24 576 .49 1.97 2

3 01000 8 64 .06 .22 0

4 10011 19 361 .31 1.23 1

Sum 1170

Avg 293

Max 576

An Example - CrossoverString Pop. Pool

(w/

Crossover)

Mate Crossover Site

New Pop. Pool

X value F(x)

1 0110|1 2 4 01100 12 144

2 01100|0 1 4 11001 25 625

3 11|000 4 2 11011 27 729

4 10|011 3 2 10000 16 256

Sum 1754

Avg 439

Max 739

GA’s in Image Segmentation

• Optimization problem

• “Twiddling Knobs” Approach

• Relationship to “Many k-armed bandit” problem

Figure: Bhanu, Lee

GA method for image segmentation

Figure: Bhanu, Lee

Images differ in characteristics such as brightness, saturation, skewness, entropy, etc.

Use these values as inputs to genetic algorithm

Figure: Bhanu, Lee

Image Analysis

Evolution of Segmentation System

Figure: Bhanu, Lee

`

The project: Implementation in DSP / FPGA

Image Capture

Image Processor

Geneticoptimizer

Collator

Memory

Output

Edge DetectorsFPGA

DSP

top related