20
1 Interactive Heuristic Edge Detection Douglas A. Lyon Computer Engineering Department Fairfield University [email protected], http://www.DocJava.com Copyright 2002 © DocJava, Inc.

1 Interactive Heuristic Edge Detection Douglas A. Lyon Computer Engineering Department Fairfield University Copyright

Embed Size (px)

DESCRIPTION

3 The Problem Given an image and points on good edges. Find a way to connect the points on the edges.

Citation preview

Page 1: 1 Interactive Heuristic Edge Detection Douglas A. Lyon Computer Engineering Department Fairfield University  Copyright

1

Interactive Heuristic Edge DetectionDouglas A. Lyon

Computer Engineering DepartmentFairfield University

[email protected], http://www.DocJava.com

Copyright 2002 © DocJava, Inc.

Page 2: 1 Interactive Heuristic Edge Detection Douglas A. Lyon Computer Engineering Department Fairfield University  Copyright

2

Background

• Good Edge Detection is hard.

• We know a good edge when we see it!

• How do we know?

• Lets assume that we do!

Page 3: 1 Interactive Heuristic Edge Detection Douglas A. Lyon Computer Engineering Department Fairfield University  Copyright

3

The Problem

• Given an image and points on good edges.

• Find a way to connect the points on the edges.

Page 4: 1 Interactive Heuristic Edge Detection Douglas A. Lyon Computer Engineering Department Fairfield University  Copyright

4

Motivation

• Experts can find good edges

• Most edge detectors are not as good

• knowledge is hard to encode.

Page 5: 1 Interactive Heuristic Edge Detection Douglas A. Lyon Computer Engineering Department Fairfield University  Copyright

5

Application

• Photo interpretation

• Path planning (source+destination)

• Medical imaging

Page 6: 1 Interactive Heuristic Edge Detection Douglas A. Lyon Computer Engineering Department Fairfield University  Copyright

6

Approach

• Pixels=graph nodes

• mark an important edge

• Search in pixel space using a heuristic

Page 7: 1 Interactive Heuristic Edge Detection Douglas A. Lyon Computer Engineering Department Fairfield University  Copyright

7

Assumptions

• A good edge is more important than speed

• Heuristics are available

• User is a domain expert

• Knowledge rep=heuristics+markers

Page 8: 1 Interactive Heuristic Edge Detection Douglas A. Lyon Computer Engineering Department Fairfield University  Copyright

8

Photo Interpretation

Page 9: 1 Interactive Heuristic Edge Detection Douglas A. Lyon Computer Engineering Department Fairfield University  Copyright

9

Echocardiogram

•3D-multi-scale analysis

Page 10: 1 Interactive Heuristic Edge Detection Douglas A. Lyon Computer Engineering Department Fairfield University  Copyright

10

Path Plans, the idea

Page 11: 1 Interactive Heuristic Edge Detection Douglas A. Lyon Computer Engineering Department Fairfield University  Copyright

11

Path Planning-pre proc.•UNAHE+thresh

•Dil+Skel

Page 12: 1 Interactive Heuristic Edge Detection Douglas A. Lyon Computer Engineering Department Fairfield University  Copyright

12

Path Planning - Search

Page 13: 1 Interactive Heuristic Edge Detection Douglas A. Lyon Computer Engineering Department Fairfield University  Copyright

13

Summary

• Heuristics+markers= knowledge

• Low-level image processing still needed

• Global optimization is not appropriate for all applications

• Sometimes we only want a single, good edge

Page 14: 1 Interactive Heuristic Edge Detection Douglas A. Lyon Computer Engineering Department Fairfield University  Copyright

14

Future work

• Teach the edge detector– Training sets?

– Neural Nets?

– Better justification for heuristics…

Page 15: 1 Interactive Heuristic Edge Detection Douglas A. Lyon Computer Engineering Department Fairfield University  Copyright

15

Open Questions

• Does the mind use open-loop filtering?

• Does the mind select from filter banks?

• Does the mind tune the filters?

Page 16: 1 Interactive Heuristic Edge Detection Douglas A. Lyon Computer Engineering Department Fairfield University  Copyright

16

Gabor filter w/threshold

• The Strong edge is the wrong edge!

Page 17: 1 Interactive Heuristic Edge Detection Douglas A. Lyon Computer Engineering Department Fairfield University  Copyright

17

Sub bands for 3x3 Robinson

Page 18: 1 Interactive Heuristic Edge Detection Douglas A. Lyon Computer Engineering Department Fairfield University  Copyright

18

Sub Bands 7x7 Gabor

Page 19: 1 Interactive Heuristic Edge Detection Douglas A. Lyon Computer Engineering Department Fairfield University  Copyright

19

Gabor-biologically motivated

Δθ =150

Page 20: 1 Interactive Heuristic Edge Detection Douglas A. Lyon Computer Engineering Department Fairfield University  Copyright

20

http://www.docjava.com