an enhanced cellular automata and image pyramid decomposition based algorithm for image segmentation...
DESCRIPTION
Two Dimensional Cellular Automata CSECS 2015 Boston3 A (bi-directional, deterministic) cellular automaton is a triple A = (S;N;δ), Where, S is a non-empty finite set of states, N is the neighborhood system, δ : S N →S is the local transition function(rule).TRANSCRIPT
An Enhanced Cellular Automata and Image Pyramid Decomposition Based Algorithm for Image Segmentation : A New Concept
Anand Prakash Shukla Suneeta Agarwal
Paper id: 40
CSECS 2015 Boston 2
Agenda
• Introduction
• The tool or technique
• Methodology
• Experiments and Results
• Conclusion
CSECS 2015 Boston 3
Two Dimensional Cellular AutomataA (bi-directional, deterministic) cellular automaton is a
triple A = (S;N; ), δWhere, S is a non-empty finite set of states, N is the neighborhood system, : Sδ N→S is the local transition function(rule).
CSECS 2015 Boston 4
Motivation
Cellular automata have a number of advantages over traditional methods of computations
• few simple rules, combination leads to more sophisticated emergent global behavior.
• Simplicity of implementation and complexity of behavior.
• CA are both inherently parallel and computationally simple.
• CA are extensible;
• Supports n-dimensions and m-label categories
• Number of labels does not increase computational time or complexity.
CSECS 2015 _ Boston 5
CA based Algorithm for Image Segmentation
The cell state Sp used here is actually a triplet (lp;θp; Cp) where lp is the label of the current cell,
θp is the strength of the current cell ,
Cp is the feature vector defined by the image.
Without loss of generality we will assume
θp ϵ [0,1]
Automata evolution rule
1
1
( )
t tp p
t tp p
p P
l l
q N P
2. (|| || . t tp q q pif g C C
1
2(|| || .
t tp q
t tp p q q
l l
g C C
CSECS 2015 _ Boston 6
Gaussian Pyramid
•Let g0 be the input image and g1 be the image obtained by filtering the original image by low pass filter.
•g1 is called reduced version of g0
•Similarly find sequence of images g0,
g1… gn
•This sequence is called Gaussian pyramid.
CSECS 2015 _ Boston 7
Laplacian and Gaussian Pyramid Generation
CSECS 2015 _ Boston 8
Proposed Method
•Select the input image I.
•Apply the Gaussian pyramid decomposition method to obtain the plane of desired level Il.
•Apply GrowCut algorithm to Il.
•Find the segmented image Il’
• Resize Il’ to original size by pyramid
reconstruction method.
CSECS 2015 _ Boston 9
Results
CSECS 2015 _ Boston 10
Segmented and reconstructed at level 2Segmented at level 0 Pixels replaced by original
image
CSECS 2015 _ Boston 11
Conclusion
.•A new approach has been proposed to improve the performance of GrowCut algorithm.
•The time taken in the segmentation decreases drastically.
•Segmentation quality is also good.
•Some background pixels are acquired at higher level of pyramid plane.