A new feature-based wavelet completed local ternary pattern (FEAT-WCLTP) for texture and medical image classification
Nowadays, texture image descriptors are used in many important real-life applications. The use of texture analysis in texture and medical image classification has attracted considerable attention. Local Binary Patterns (LBP) is one of the simplest yet eff ective texture descriptors. But it has some...
Saved in:
Main Author: | |
---|---|
Format: | Thesis |
Language: | English |
Published: |
2019
|
Subjects: | |
Online Access: | http://umpir.ump.edu.my/id/eprint/32507/1/A%20new%20feature-based%20wavelet%20completed%20local%20ternary%20pattern%20%28FEAT-WCLTP%29%20for%20texture%20and%20medical%20image%20classification.wm.pdf http://umpir.ump.edu.my/id/eprint/32507/ |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
id |
my.ump.umpir.32507 |
---|---|
record_format |
eprints |
spelling |
my.ump.umpir.325072023-01-25T01:38:59Z http://umpir.ump.edu.my/id/eprint/32507/ A new feature-based wavelet completed local ternary pattern (FEAT-WCLTP) for texture and medical image classification Shamaileh, Abeer Moh'd Salem QA75 Electronic computers. Computer science Nowadays, texture image descriptors are used in many important real-life applications. The use of texture analysis in texture and medical image classification has attracted considerable attention. Local Binary Patterns (LBP) is one of the simplest yet eff ective texture descriptors. But it has some limitations that may affect its accuracy. Hence, different variants of LBP were proposed to overcome LBP’s drawbacks and enhance its classification accuracy. Completed local ternary pattern (CLTP) is one of the significant LBP variants. However, CLTP suffers from two main limitations: the selection of the threshold value is manually based and the high dimensionality which is negatively affected the descriptor performance and leads to high computations. This research aims to improve the classification accuracy of CLTP and overcome the computational limitation by proposing new descriptors inspired by CLTP. Therefore, this research introduces two contributions: The first one is a proposed new descriptor that integrates redundant discrete wavelet transform (RDWT) with the original CLTP, namely, wavelet completed local ternary pattern (WCLTP). Extracting CLTP in wavelet transform will help increase the classification accuracy due to the shift invariant property of RDWT. Firstly, the image is decomposed into four sub-bands (LL, LH, HL, HH) by using RDWT. Then, CLTP is extracted based on the LL wavelet coefficients. The latter one is the reduction in the dimensionality of WCLTP by reducing its size and a proposed new texture descriptor, namely, feature-based wavelet completed local ternary pattern (FeatWCLTP). The proposed Feat-WCLTP can enhance CLTP’s performance and reduce high dimensionality. The mean and variance of the values of the selected texture pattern are used instead of the normal magnitude texture descriptor of CLTP. The performance of the proposed WCLTP and Feat-WCLTP was evaluated using four textures (i.e. OuTex, CUReT, UIUC and Kylberg) and two medical (i.e. 2D HeLa and Breast Cancer) datasets then compared with several well-known LBP variants. The proposed WCLTP outperformed the previous descriptors and achieved the highest classification accuracy in all experiments. The results for the texture dataset are 99.35% in OuTex, 96.57% in CUReT, 94.80% in UIUC and 99.88% in the Kylberg dataset. The results for the medical dataset are 84.19% in the 2D HeLa dataset and 92.14% in the Breast Cancer dataset. The proposed Feat-WCLTP not only overcomes the dimensionality problem but also considerably improves the classification accuracy. The results for Feat-WCLTP for texture dataset are 99.66% in OuTex, 96.89% in CUReT, 95.23% in UIUC and 99.92% in the Kylberg dataset. The results for the medical dataset are 84.42% in the 2D HeLa dataset and 89.12% in the Breast Cancer dataset. Moreover, the proposed Feat-WCLTP reduces the size of the feature vector for texture pattern (1,8) to 160 bins instead of 400 bins in WCLTP. The proposed WCLTP and Feat-WCLTP have better classification accuracy and dimensionality than the original CLTP. 2019-08 Thesis NonPeerReviewed pdf en http://umpir.ump.edu.my/id/eprint/32507/1/A%20new%20feature-based%20wavelet%20completed%20local%20ternary%20pattern%20%28FEAT-WCLTP%29%20for%20texture%20and%20medical%20image%20classification.wm.pdf Shamaileh, Abeer Moh'd Salem (2019) A new feature-based wavelet completed local ternary pattern (FEAT-WCLTP) for texture and medical image classification. Masters thesis, Universiti Malaysia Pahang. |
institution |
Universiti Malaysia Pahang |
building |
UMP Library |
collection |
Institutional Repository |
continent |
Asia |
country |
Malaysia |
content_provider |
Universiti Malaysia Pahang |
content_source |
UMP Institutional Repository |
url_provider |
http://umpir.ump.edu.my/ |
language |
English |
topic |
QA75 Electronic computers. Computer science |
spellingShingle |
QA75 Electronic computers. Computer science Shamaileh, Abeer Moh'd Salem A new feature-based wavelet completed local ternary pattern (FEAT-WCLTP) for texture and medical image classification |
description |
Nowadays, texture image descriptors are used in many important real-life applications. The use of texture analysis in texture and medical image classification has attracted considerable attention. Local Binary Patterns (LBP) is one of the simplest yet eff ective texture descriptors. But it has some limitations that may affect its accuracy. Hence, different variants of LBP were proposed to overcome LBP’s drawbacks and enhance its classification accuracy. Completed local ternary pattern (CLTP) is one of the significant LBP variants. However, CLTP suffers from two main limitations: the selection of the threshold value is manually based and the high dimensionality which is negatively affected the descriptor performance and leads to high computations. This research aims to improve the classification accuracy of CLTP and overcome the computational limitation by proposing new descriptors inspired by CLTP. Therefore, this research introduces two contributions: The first one is a proposed new descriptor that integrates redundant discrete wavelet transform (RDWT) with the original CLTP, namely, wavelet completed local ternary pattern (WCLTP). Extracting CLTP in wavelet transform will help increase the classification accuracy due to the shift invariant property of RDWT. Firstly, the image is decomposed into four sub-bands (LL, LH, HL, HH) by using RDWT. Then, CLTP is extracted based on the LL wavelet coefficients. The latter one is the reduction in the dimensionality of WCLTP by reducing its size and a proposed new texture descriptor, namely, feature-based wavelet completed local ternary pattern (FeatWCLTP). The proposed Feat-WCLTP can enhance CLTP’s performance and reduce high dimensionality. The mean and variance of the values of the selected texture pattern are used instead of the normal magnitude texture descriptor of CLTP. The performance of the proposed WCLTP and Feat-WCLTP was evaluated using four textures (i.e. OuTex, CUReT, UIUC and Kylberg) and two medical (i.e. 2D HeLa and Breast Cancer) datasets then compared with several well-known LBP variants. The proposed WCLTP outperformed the previous descriptors and achieved the highest classification accuracy in all experiments. The results for the texture dataset are 99.35% in OuTex, 96.57% in CUReT, 94.80% in UIUC and 99.88% in the Kylberg dataset. The results for the medical dataset are 84.19% in the 2D HeLa dataset and 92.14% in the Breast Cancer dataset. The proposed Feat-WCLTP not only overcomes the dimensionality problem but also considerably improves the classification accuracy. The results for Feat-WCLTP for texture dataset are 99.66% in OuTex, 96.89% in CUReT, 95.23% in UIUC and 99.92% in the Kylberg dataset. The results for the medical dataset are 84.42% in the 2D HeLa dataset and 89.12% in the Breast Cancer dataset. Moreover, the proposed Feat-WCLTP reduces the size of the feature vector for texture pattern (1,8) to 160 bins instead of 400 bins in WCLTP. The proposed WCLTP and Feat-WCLTP have better classification accuracy and dimensionality than the original CLTP. |
format |
Thesis |
author |
Shamaileh, Abeer Moh'd Salem |
author_facet |
Shamaileh, Abeer Moh'd Salem |
author_sort |
Shamaileh, Abeer Moh'd Salem |
title |
A new feature-based wavelet completed local ternary pattern (FEAT-WCLTP) for texture and medical image classification |
title_short |
A new feature-based wavelet completed local ternary pattern (FEAT-WCLTP) for texture and medical image classification |
title_full |
A new feature-based wavelet completed local ternary pattern (FEAT-WCLTP) for texture and medical image classification |
title_fullStr |
A new feature-based wavelet completed local ternary pattern (FEAT-WCLTP) for texture and medical image classification |
title_full_unstemmed |
A new feature-based wavelet completed local ternary pattern (FEAT-WCLTP) for texture and medical image classification |
title_sort |
new feature-based wavelet completed local ternary pattern (feat-wcltp) for texture and medical image classification |
publishDate |
2019 |
url |
http://umpir.ump.edu.my/id/eprint/32507/1/A%20new%20feature-based%20wavelet%20completed%20local%20ternary%20pattern%20%28FEAT-WCLTP%29%20for%20texture%20and%20medical%20image%20classification.wm.pdf http://umpir.ump.edu.my/id/eprint/32507/ |
_version_ |
1756060200482111488 |
score |
13.18916 |