Development of a Free Image Processing Toolbox for Scilab

1
Development of a Free Image Processing Toolbox
for Scilab
Author Ricardo Fabbri Advisor
Luciano da Fontoura Costa e-mail
rfabbri_at_if.sc.usp.br URL
siptoolbox.sourceforge.net Instituto de Ciências
Matemáticas e de Computação Universidade de São
Paulo, Brasil.
09/2003
2
Plan of Presentation

• Introduction
• Design and Development
• Implemented Functionalities
• Future Work
• The Cybernetic Vision Research Group
• Conclusions
• References

3
Introduction
4
SciLab

• Free software
• Numerical programming
• Rapid Prototyping
• Extensive libraries and toolboxes
• Easy for matrix manipulation
• A B C D
• Unnecessary looping

5
SciLab

• Ambiente open source
• Programacao numerica
• Prototipacao
• Rotinas prontas
• Facil para manipular matrizes
• A B C D
• Loop desnecessario

Ideal for Image Processing
6
SIP toolbox
Goals

• Image I/O for SciLab
• Image processing, analysis, and recognition.
• Free software
• Cross-platform

7
SIP toolbox

• Beginning only a few simple routines
• By the end of 2001
• Routines for image I/O in virtually any format
• Many other routines for IP
• Today it is used by many researchers and students
  around the globe

8
Development
9
Software Design
10
Auto-configuration
Problems with portability

• Differences between development and usage
  environments
• UNIX has many variants and is highly customizable
• How to detect external libraries?
• How to deal with different versions of Scilab?

11
GNU Autotools

• Autoconf, Automake, and Libtool
• Typical in free software
• Enable installation through the famous sequence
• configure tests the system
• Make compiles
• Make install installs

12
Autotoconf and Automake
13
My Experience with Free Development

• Release early, Release Often
• FEEDBACK
• Linux/UNIX
• Sourceforge.net
• Requirements
• Effective user inclusion

14
Free Development

• Architectural design (high level)
• SIP is simple
• Joint usage of routines is not enforced by the
  software policy
• Detailed design
• Interface Matlab
• Good sourcecode organization

15
Free Development

• Test
• Interactive prototyping
• Alpha and beta versions
• Main developer tests less and fixes more.
• Deployment
• Freshmeat.net and Sourceforge.net
• email and newgroups

16
Functionalities
17
Image I/O

• Read image files
• imread
• Write imag files
• imwrite
• Almost 80 formats JPEG, TIFF, BMP, PNG, GIF,
  FITS, XPM, Photo-CD, ...
• ImageMagick

18
1D Gaussian filtering

• Convolution by Fourier Transform

19
1D Gaussian filtering
20
Multiscale Skeletons
21
Multiscale Skeletons
22
Multiscale Skeletons

• Stability under rotations
• Accepts objects with holes
• Fast algorithms based on graph theory

23
Other Functionalities

• Thinning (Lam-Suen)
• Hough Transform
• Fractal Dimension
• Edge detector by Fourier derivatives
• Watershed segmentation
• Fast euclidean distance transforms

24
2D Gaussian filter
25
2D Gaussian filtering
26
Median Filtering

• Non-linear filter with low border shifting

27
Border Enhancement

• Partial derivative approximation
• Two orthogonal directions
• Gradient magnitude

• Convolution masks

28
Border Enhancement
Sobel
ImageMagick
29
Curvature

• Useful local measure

30
Mathematical Morphology

• Dilation
• Erosion
• Arbitrary structuring elements

31
Geometric Tansformations

• Coded using ImageMagick
• Mogrify command

32
Edge follower

• Extracts parametric coordinates
• C(t) (x(t), y(t))
• Closed curves and contours

33
Histogram Equalization

• Spreads the distribution of intensities

34
Contributions

• Denoising by min/max curvature flow (thanks to L.
  Estrozi)
• Routines for processing images from optical
  interferometry (J. Druel)
• Adaptive thresholding (ImageMagick)

35
Future Work

• Better integer support
• Test suite
• Improve graphical display
• External doc for Animal

36
Future Work

• Integration with Imagemagick and Gandalf
  libraries
• CVS
• XML Help
• Usability
• New routines

37
Conclusions

• High demand for SIP
• It is still the only free tool for easy
  prototyping of image analysis solutions, not
  limited to simple processing techniques
• Low cost
• Valuable experience with free software

38
Personal Facts

• Undergrad research since 1999
• Cybernetic Vision Research Group
• Very well equipped
• Productive environment
• I'm currently working with distance transforms,
  Voronoi diagrams, skeletonization, and related
  techniques.
• Plan to do a PhD in Image Processing.

39
References