Normal view MARC view ISBD view

Parallel Genetic Algorithm based Thresholding Schemes for Image Segmentation

By: Kanungo, Priyadarshi.
Contributor(s): Nanda, Pradipta Kumar [Supervisor] | Department of Electrical Engineering.
Material type: materialTypeLabelBookPublisher: 2009Description: 199 p.Subject(s): Engineering and Technology | Electrical Engineering | Image SegmentationOnline resources: Click here to access online Dissertation note: Thesis (Ph.D)- National Institute of Technology, Rourkela Summary: In this thesis, the problem of image segmentation has been ad dressed using the notion of thresh- olding. Since the focus of this work is primarily on object/o bjects background classification and fault detection in a given scene, the segmentation problem i s viewed as a classification problem. In this regard, the notion of thresholding has been used to cl assify the range of gray values and hence classifies the image. The gray level distributions of t he original image or the proposed feature image have been used to obtain the optimal threshold . Initially, PGA based class models have been developed to cla ssify different classes of a nonlinear multimodal function. This problem is formulated where the nonlinear multimodal function is viewed as consisting of multiple class distribu tions. Each class could be represented by the niche or peaks of that class. Hence, the problem has bee n formulated to detect the peaks of the functions. PGA based clustering algorithm has b een proposed to maintain stable sub-populations in the niches and hence the peaks could be de tected. A new interconnection model has been proposed for PGA to accelerate the rate of conv ergence to the optimal solution. Convergence analysis of the proposed PGA based algorithm ha s been carried out and is shown to converge to the solution. The proposed PGA based clusteri ng algorithm could successfully be tested for different classes and is found to converge much fa ster than that of GA based clustering algorithm. Two thresholding schemes namely Feature Less (FL) and Featu re Based (FB) thresholding have been proposed using the PGA based clustering algorithm and PGA based optimization strategy. Both the approaches have been tested with images o f different classes and it has been found that FB approach proved to be better than FL approach. T he performance of the proposed approaches are found to be better than Otsu’s and Kwon’s meth ods in many cases. A Minimum Mean Square Error (MMSE) based FL and FB schemes hav e been proposed to deal with fault detection in a given scene whose histogram does not exhibit clear bi-modality and almost becomes unimodal. These schemes also employ the p roposed PGA based clustering iii algorithm. The schemes could successfully be tested with im ages of earth surface cracks and performance of the proposed method proved to be better than F uang’s fault detection method. The scheme could also be validated with general images and th e efficacy has been demonstrated especially with image for colour-blindness. Adaptive thresholding based schemes have been proposed to s eparate object and back- ground in images with nonuniform lighting conditions. The m ethods are based on the notion of window merging and window growing. Three new window selec tion criteria have been pro- posed to adaptively fix the size of windows for segmentation. The selected windows have been segmented by Otsu’s, Kwon’s, the proposed PGA, and MMSE base d schemes. Sizes of the windows have also been fixed based on the window growing appro ach where, selection of win- dows is based on notion of entropy and feature entropy. The wi ndows, thus fixed, have been segmented by Otsu’s, Kwon’s, and MMSE based approaches. The results obtained by window merging and window growing are found to be better than that of results obtained by Huang’s approach. The efficacy of the proposed schemes has been demon strated with different images of having nonuniform lighting condition
Tags from this library: No tags from this library for this title. Log in to add tags.
    average rating: 0.0 (0 votes)
Item type Current location Collection Call number Status Date due Barcode
Thesis (Ph.D/M.Tech R) Thesis (Ph.D/M.Tech R) Thesis Section Reference Not for loan T58

Thesis (Ph.D)- National Institute of Technology, Rourkela

In this thesis, the problem of image segmentation has been ad
dressed using the notion of thresh-
olding. Since the focus of this work is primarily on object/o
bjects background classification and
fault detection in a given scene, the segmentation problem i
s viewed as a classification problem.
In this regard, the notion of thresholding has been used to cl
assify the range of gray values and
hence classifies the image. The gray level distributions of t
he original image or the proposed
feature image have been used to obtain the optimal threshold
.
Initially, PGA based class models have been developed to cla
ssify different classes of a
nonlinear multimodal function. This problem is formulated
where the nonlinear multimodal
function is viewed as consisting of multiple class distribu
tions. Each class could be represented
by the niche or peaks of that class. Hence, the problem has bee
n formulated to detect the
peaks of the functions. PGA based clustering algorithm has b
een proposed to maintain stable
sub-populations in the niches and hence the peaks could be de
tected. A new interconnection
model has been proposed for PGA to accelerate the rate of conv
ergence to the optimal solution.
Convergence analysis of the proposed PGA based algorithm ha
s been carried out and is shown
to converge to the solution. The proposed PGA based clusteri
ng algorithm could successfully be
tested for different classes and is found to converge much fa
ster than that of GA based clustering
algorithm.
Two thresholding schemes namely Feature Less (FL) and Featu
re Based (FB) thresholding
have been proposed using the PGA based clustering algorithm
and PGA based optimization
strategy. Both the approaches have been tested with images o
f different classes and it has been
found that FB approach proved to be better than FL approach. T
he performance of the proposed
approaches are found to be better than Otsu’s and Kwon’s meth
ods in many cases.
A Minimum Mean Square Error (MMSE) based FL and FB schemes hav
e been proposed
to deal with fault detection in a given scene whose histogram
does not exhibit clear bi-modality
and almost becomes unimodal. These schemes also employ the p
roposed PGA based clustering
iii
algorithm. The schemes could successfully be tested with im
ages of earth surface cracks and
performance of the proposed method proved to be better than F
uang’s fault detection method.
The scheme could also be validated with general images and th
e efficacy has been demonstrated
especially with image for colour-blindness.
Adaptive thresholding based schemes have been proposed to s
eparate object and back-
ground in images with nonuniform lighting conditions. The m
ethods are based on the notion
of window merging and window growing. Three new window selec
tion criteria have been pro-
posed to adaptively fix the size of windows for segmentation.
The selected windows have been
segmented by Otsu’s, Kwon’s, the proposed PGA, and MMSE base
d schemes. Sizes of the
windows have also been fixed based on the window growing appro
ach where, selection of win-
dows is based on notion of entropy and feature entropy. The wi
ndows, thus fixed, have been
segmented by Otsu’s, Kwon’s, and MMSE based approaches. The
results obtained by window
merging and window growing are found to be better than that of
results obtained by Huang’s
approach. The efficacy of the proposed schemes has been demon
strated with different images
of having nonuniform lighting condition

There are no comments for this item.

Log in to your account to post a comment.


Implemented and Maintained by Biju Patnaik Central Library.
For any Suggestions/Query Contact to library or Email: library@nitrkl.ac.in OR bpcl-cir@nitrkl.ac.in. Ph:91+6612462103
Website/OPAC best viewed in Mozilla Browser in 1366X768 Resolution.

Powered by Koha