Image Processing Reference
In-Depth Information
9.3
Cellular Automaton Interactive Segmentation
In an intuitive user interaction scheme - user specifies certain image pixels (call
them seed pixels) that belong to objects, should be segmented from each other. The
task is to assign labels to all other image pixels automatically, preferably achieving
the segmentation result expecting to get. The task statement and input data is similar
to Graph Cuts in [3] and [11], however the segmentation instrument differs.
Interactive segmentation method based on CA is used for solving cell labeling
task, which allows (but not requires) human input during labeling process, to provide
dynamic interaction and feedback between the user and the algorithm. Important
properties of CA method, that would like to outline are:
1. Capable of solving moderately hard segmentation tasks (see examples in this
chapter);
2. Works with images of any dimension N
1;
3. Performs multi-label image segmentation (the computation time is not directly
affected by the number of labels);
4. Is extensible, allowing construction of new families of segmentation algo-
rithms with specific properties;
5. Interactivity - as the segmentation is refined with each iteration, user can ob-
serve the evolution and refine the segmentation 'on the fly';
6. The algorithm is simple in both understanding and implementation;
7. Using cellular automata allows fast parallel implementation
The proposed method iteratively gives feedback to the user while evolution. This
process allows user guidance of the implementation, yet does not require additional
efforts where the segmentation is reliably computed.
9.3.1
Labels in Image Plane
A digital image is a two-dimensional array of k
m pixels. An unlabeled image can
be considered as a particular configuration state of a CA, where cellular space P is
defined by the k
×
×
m array set of the image, and initial states for
p
P are set to:
, C p =
l p =
0
, θ p =
0
RGB p
(9.4)
where RGB p is the three dimensional vector of pixel p 's color in RGB space. The
final goal of the segmentation is to assign each pixel with one of the K possible
labels.
We can treat pixel labeling process as growth and struggle for domination of K
types of bacteria. The bacteria start to spread (grow) from the seed pixels and try to
occupy all the image. The rules of bacteria growth and competition are obvious -
at each discrete time step, each cell tries to 'attack' its neighbors. The attack force
is defined by the attacker cell's strength
θ q , and the distance between attacker's and
defender's feature vectors C q and
C p . If the attack force is greater than defender's
strength - the defending cell is 'conquered' and its label and strength are changed.
Search WWH ::




Custom Search