Authors: Andrew Blake and Michael Isard


Active Contours is about the computer analysis of moving video images. It develops geometric and probabilistic models  for shapes and their dynamics.  The models are applied to the real-time analysis of shapes in motion, and addresses issues of learning, temporal filtering and the problems of visual clutter. Numerous applications are illustrated from computer graphics animation, user-interface design, medical imaging, automated surveillance and robotics.


Full text now downloadable in postscript (zipped) and in pdf:

   Preface, contents, glossary, index  ps (0.3Mb)  and pdf (0.2 Mb)

   Introductory section ps (6.3 Mb)  and pdf (1.1 Mb)

   Section 1 -  Geometric fundamentals ps (4.3Mb)  and pdf (1.1 Mb)

   Section 2  - Probabilistic modelling ps (4.7 Mb)  and pdf (1.7 Mb)

   Appendix  ps (0.3 Mb)  and pdf (0.2 Mb)



Additional material on the web includes background information on dynamical analysis of visual motion, including MPEG motion sequences and research papers.


Errata for the book are available here and will be updated as errors are discovered.


Contents

Approx. 350 pages, and lots of pictures, many in colour.

Foreword by Professor Demetri Terzopoulos, University of Toronto.

1. Introduction: organisation of the book, applications.

2. Active shape models: snakes, deformable templates, dynamic contours.

PART I: Geometrical Fundamentals

3. Spline curves: B-spline functions, finite bases, multiple knots, norm and inner product for spline functions, B-spline parametric curves, curves with vertices, control vector, norm for curves, areas and moments.

4. Shape-space models: representing transformations in shape-space, the space of Euclidean similarities, planar affine shape-space, norms and moments in a shape-space, perspective and weak perspective, three-dimensional affine shape-space, key-frames, articulated motion.

5. Image processing techniques for feature location: linear scanning, image filtering, using colour, correlation matching, background subtraction.

6. Fitting spline templates: regularised matching, normal displacement in curve fitting, recursive solution of curve-fitting problems, examples.

7. Pose recovery: calculating the pose of a planar object, pose recovery for three-dimensional objects, separation of rigid and non-rigid motion.

PART II: Probabilistic Modelling

8: Probabilistic models of shape: probability distributions over curves, posterior distribution, probabilistic modelling of image features, validation gate, learning the prior, Principal Components Analysis (PCA).

9. Dynamical models: some simple dynamical prior distributions, first-order auto-regressive processes, limitations of first-order dynamical models, second-order dynamical models, second-order AR processes in shape-space, setting dynamical parameters.

10. Dynamic contour tracking: temporal fusion by Kalman filter tracking performance, choosing dynamical parameters, case study.

11. Learning motion: learning one-dimensional dynamics, learning AR process dynamics in shape-space, dynamical modes, performance of trained trackers.

12. Non-Gaussian models and random sampling algorithms: factored sampling, the Condensation algorithm, an observation model for Condensation, applications of the Condensation algorithm.

APPENDIX

A. Mathematical background: vectors and matrices, B-spline basis functions, probability.

B. Stochastic dynamical systems: continuous-time first-order dynamics, second-order dynamics in continuous time, accuracy of learning.

C. Further shape-space models: recursive synthesis of shape-spaces.

Glossary of notation, bibliography, author index, index.