Deformable/Active Contours (or Snakes)

(Trucco, Chapt 4)

- The goal is to find a contour that best approximates the perimeter of an object.

- It is helpful to visualize it as a rubber band of arbitrary shape that is capable of

- It is applied to the gradient magnitude of the image, not to the edge points (e.g.,
like the Hough transform).

• Procedure

- Snakes do not solve the entire problem of finding contours in images.

- They depend on other mechanisms such as interaction with a user or with some
other higher-level computer vision mechanism:

(1) First, the snake is placed near the image contour of interest.

(2) During an iterative process, the snake is attracted towards the target con-
tour by various forces that control the shape and location of the snake within
the image.
 -2-

• Approach

- It is based on constructing an energy functional which measures the appropriate-

- Good solutions correspond to minima of the functional.

- The goal is to minimize this functional with respect to the contour parameters.

• Contour parameterization

- The snake is a contour represented parametrically as c(s) = (x(s), y(s)) where

• The energy functional

- The energy functional used is a sum of several terms, each corresponding to some
force acting on the contour.

- A suitable energy functional is the sum the following three terms:

E = ∫ ( (s)E cont + (s)E curv + (s)E image )ds

- The parameters , , and control the relative influence of the corresponding

energy terms and can vary along c.

• Interpretation of the functional’s terms

- Each energy term serves a different purpose:

E image : it attracts the contour toward the closest image edge.

E cont : it forces the contour to be continuous.

E curv : it forces the contour to be smooth.

- E cont and E curv are called internal energy terms.

- E image is called external energy term.

• The continuity term

- Minimize the first derivative:

dc 2
E cont = || ||
ds
- In the discrete case, the contour is approximated by N points p1 , p2 , . . . , p N and
the first derivative is approximated by a finite difference:

E cont = || pi − pi−1 ||2 or

E cont = (x i − x i−1 )2 + (y i − y i−1 )2

- This term tries to minimize the distance between the points, however, it has the
effect of causing the contour to shrink.

- A better form for E cont is the following:

E cont = (d − || pi − pi−1 ||)2

where d is the average distance between the points of the snake.

- The new E cont attempts to keep the points at equal distances (i.e, spread them
equally along the snake).
 -4-

• The smoothness term

- The purpose of this term is to enforce smoothness and avoid oscillations of the
snake by penalizing high contour curvatures.

- Minimize the second derivative (curvature):

d2c 2
E curv = || 2 ||
ds

- In the discrete case, the curvature can be approximated by the following finite dif-
ference:
E curv = || pi−1 − 2 pi + pi+1 ||2 or

E curv = (x i−1 − 2x i + x i+1 )2 + (y i−1 − 2y i + y i+1 )2

• The edge attraction term

- The purpose of this term is to attract the contour toward the target contour.

- This can be achieved by the following function:

E image = − ||∇I ||

where ∇I is the gradient of the intensity computed at each snake point.

- Note that E image becomes very small when the snake points get close to an edge.
 -5-

• Discrete formulation of the problem

Assumptions

Let I be an image and p1 , ..., p N the initial locations of the snake (evenly spaced,
chosen close to the contour of interest).

Problem Statement

Starting from p1 , ..., p N , find the deformable contour p1 , ..., p N which fits the tar-
get contour by minimizing the energy functional:

N
Σ(
i=1


i E cont + 

i E curv + 

i E image )

• A greedy algorithm

- A greedy algorithm makes locally optimal choices, hoping that the final solution
will be globally optimum.

Step1 (greedy minimization): each point of the snake is moved within a small
neighborhood (e.g., MxM) to the point which minimizes the energy functional

Step 2 (corner elimination): search for corners (curvature extrema) along the con-
tour; if a corner is found at point p j , set j to zero.
 -6-

Algorithm
The input is an intensity image I containing the target contour and points
p1 ,..., p N , defining the initial position and shape of the snake.

1. For each pi , i = 1, . . , N , search its MxM neighborhood to find the location

2. Estimate the curvature of the snake at each point and look for local maxima
(i.e., corners); Set j to zero for each p j at which the curvature is a local max-


imun and exceeds a threshold.

3. Update the value of d.

Repeat steps 1-3 until only a very small fraction of snake points move in an iter-
ation.
 -7-

• Implementation details

- It is important to normalize the contribution or each term for correct implementa-

(1) For E cont and E curv , it is sufficient to divide by the largest value in the
neighborhood in which the point can move.

||∇I || − min
(2) normalize ||∇I || as where min and max are the minimum and
max − min
maximum gardient values in the neighborhood.

• Comments

- This approach is simple and has low computational requirements (O(MN ).

- It does not guarantee convergence to the global minimum of the functional.

- Works very well as far as the initial snake is not too far from the desired solution.