2-D edge detection using snakes

1
2-D edge detection using snakes

• Project 13 Team 6

2
Snakes in general

• Snakes or active contours are computer-generated
  curves that move within images to find object
  boundaries.

3
Snakes in general

• Internal forces (snake attribute)
• Prevent streching
• Prevent bending
• External forces (picture attribute)
• Guide towards the object boundaries

4
Traditional snake

• Needs to be initalized close to the object to
  find it
• Can not go into concavities

5
Gradient Vector Flow (GVF) snake

• Can go into concavities
• Can be initialized anywhere, even across
  boundaries

6
Gradient Vector Flow (GVF) snake
7
The team

• Renáta Hodován assistance, QA
• Ferenc Nasztanovics homepage
• Julian Stottinger MATLAB
• István Vincze C
• Péter Siklós presentation

8
Original plan

• Found a semi working GVF snake in MATLAB
  (original program by Dejan Tomazevic)
• Had a working C GUI but without any force
  powers (by István Vincze)

Extract the snake force features from the MATLAB
program, and reimplement in C, then add some
extra fetures
9
The C program

• Developement of the internal snake energies
• Improvements in GUI functionality
• Iteration and snake movement
• Calculation of snake direction
• Optimalization of snake energy
• Tuning internal energies
• Embedding the GVF funtcion

10
Plan B

• C program not coming along, so
• Try some features in the MATLAB program
• Both programs use the same function to calculate
  GVF, so results shouldnt be so different
• Save some time experimenting in C

11
MATLAB experimentation

• Wanted to do a snake finding the edge of the team
  on a group photo

12
MATLAB experimentation

• Some holes were not accessible

13
MATLAB experimentation

• Reproduced the problem on a smaller image

14
MATLAB experimentation

• Conclusion
• This problem can not be solved with the
  traditional GVF snake
• Some new energies need to be introduced, to force
  the snake into these kind of cavities.
• Readjust snake using tracing quadtrees (see
  presentation on thermal video processing by
  András Hajdu)
• Contour optimization
• Snake area minimalization?

15
MATLAB experimentation

• Importance of parameter tuning

Small blur
High blur
16
C program results
Lung

• Team photo

This is an ugly formula
17
Thank you for your attention!