First and second masks are obtained

First and second masks are obtained by placing two different thresholds on the horizontal , vertical, and
diagonal components. Two different thresholds are used in this approach to get more edge information.
x Third mask is obtained by looking at the maximum pixel value of the horizontal , vertical, and diagonal
components.
x Forth mask is obtained by finding the max intensity pixels among h, v and d components and by multiplying
with approximation component.
Steps of Edge detection with wavelet that show the details of the feature extraction are describes as follows:
Asmaa Hashem Sweidan et al. / Procedia Computer Science 65 ( 2015 ) 601 – 611 607
1. After pre-processing phase Decompose the image using wavelets.
2. Get approximation a, horizontal y, vertical v and diagonal d components.
3. Filter out the strong edges of horizontal, vertical and diagonal components by using T1ߛ and T2ߛ
where T is threshold and ߛ is the standard deviation of respected image.
4. Obtain first edge map by applying T1ߛ on h, v and d components and combining them and then
multiplying the resulting mask with the approximation component.
5. Repeat step5 on T2ߛ
6. Repeat step5 on T1ߛ and T2ߛ
7. Repeat step5 on finding the max intensity pixels among h, v and d components and by multiplying
with approximation component
In this paper moment invariants are used to represent a shape . Feature vectors are calculated for input image
and image data base. shape feature vector are given by equations (5-11).
߶ሺͳሻ ൌ ߟଶ଴ ൅ ߟ଴ଶ (5)

߶ሺʹሻ ൌ ሺߟଶ଴ െ ߟ଴ଶሻଶ ൅ Ͷߟଵଵ
ଶ (6)

߶ሺ͵ሻ ൌ ሺߟଷ଴ െ ͵ߟଵଶሻଶ ൅ ሺ͵ߟଶଵ െ ߟ଴ଷሻଶ (7)

߶ሺͶሻ ൌ ሺߟଷ଴ ൅ ߟଵଶሻଶ ൅ ሺߟଶଵ ൅ ߟ଴ଷሻଶ (8)
߶ሺͷሻ ൌ ሺߟଷ଴ െ ͵ߟଵଶሻሺߟଷ଴ ൅ ߟଵଶሻሾሺߟଷ଴ ൅ ߟଵଶሻଶ െ ͵ሺߟଶଵ ൅ ߟ଴ଷሻଶሿ ൅ ሺ͵ߟଶଵ െ ߟ଴ଷሻሺߟଶଵ ൅ ߟ଴ଷሻሾ͵ሺߟଷ଴ ൅
ߟଵଶሻଶ െ ሺߟଶଵ ൅ ߟ଴ଷሻଶሿሺͻሻ
߶ሺ͸ሻ ൌ ሺߟଶ଴ െ ߟ଴ଶሻሾሺߟଷ଴ ൅ ߟଵଶሻଶ െ ሺߟଶଵ ൅ ߟ଴ଷሻଶሿ ൅ Ͷߟଵଵሺߟଷ଴ ൅ ߟଵଶሻሺߟଶଵ ൅
ߟ଴ଷሻሺͳͲሻ
߶ሺ͹ሻ ൌ ሺ͵ߟଶଵ െ ߟ଴ଷሻሺߟଷ଴ ൅ ߟଵଶሻሾሺߟଷ଴ ൅ ߟଵଶሻଶ െ ͵ሺߟଶଵ ൅ ߟ଴ଷሻଶሿ െ ሺߟଷ଴ െ ͵ߟଵଶሻሺߟଶଵ ൅ ߟ଴ଷሻሾ͵ሺߟଷ଴
൅ߟଵଶሻଶ െ ሺߟଶଵ ൅ ߟ଴ଷሻଶሿሺͳͳሻ
The PCA algorithm has been utilized for feature extraction. PCA transformed space contribution to the sub-spaces
to reduce the dimensions. Then, it turns the contribution of space in sub-spaces to reduce the dimensions. A shape
feature vector will be formed as a four edge maps.
4.3. Classification phase
In this stage, SVMs classifier is used for classification of feature vectors from features extracted stage via
classifying input image using a trained model. This phase employs retrieving all the best matching images from the
matching class of the input image, whereas the outputs are the corresponding water quality degree equivalent to each
image in the testing dataset. The proposed approach addresses a multi-class problem where a variety of techniques
was used for decomposition of the multi-class problem into several binary problems using SVMs as binary classifiers.
In this research, We used one-against-all approach with 10-fold cross validation to solve a multi-class problems.
Algorithm (1) shows the details about the classification algorithm.
5. Experimental Analysis and Discussion
Nile Tilipia "Oreochromis niloticus” is pollution resistant species ideal for biomarker of water pollution. The
datasets used for experiments were constructed based on real sample microscopic images for fish gills in different
histopathlogical change stages exposed to copper and water pH.
Fish images were collected from Abbassa farm, Abo-Hammad, Sharkia Governote, Egypt. Some samples of both
training and testing datasets are shown in fig. 2. Training dataset is divided into 4 classes representing the different
608 Asmaa Hashem Sweidan et al. / Procedia Computer Science 65 ( 2015 ) 601 – 611
water quality degrees; namely excellent quality, good quality, moderate quality, and bad quality [19], as shown in fig.
3
Algorithm (1) Feature selection and classifier training
Fig.2. Examples of training and testing fish gills microscopic images
x Excellent water quality: showing primary filament (F) and secondary lamellae (L) arising from these,
parallel to each other and perpendicular to the filament axis, as shown in figure 3.a.
x Good water quality: where fish exposed to copper at pH 9 were more or less similar to those of control
group, as shown in figure 3.b.
x Moderate water quality: where fish exposed to copper at pH 7 showing hyperplasia of the primary lamellae
with complete fusion of the secondary lamellaeand shortened , as shown in figure 3.c.
x Bad water quality: where fish exposed to copper at pH 5 Drooping and shortening of some of the
secondary lamellae , as shown in figure 3.d.
Input: Training data as example ܺ଴ ൌ ሾݔͳǡ ݔʹǡ Ǥ Ǥ Ǥ ݔ݈ሿ
்
Initialize: Index for selected features: f=[1,2,...n]
Train the SVMs classifier using samples ܺ଴
For t=1,...,T do
Compute the ranking criteria according to the trained SVMs
Select the top ܯ௧features, and eliminate the other features
Restrict training examples to selected feature
Construct ܰ binary SVMs.
Each SVM separates one class from the rest classes
Train the ݅
௧௛ SVM with all training samples of the ݅
௧௛ class with positive
labels, and training samples of other classes with negative labels
End for
Output: Classify the water quality degree
Asmaa Hashem Sweidan et al. / Procedia Computer Science 65 ( 2015 ) 601 – 611 609
Fig. 3. Examples of water quality degrees
The proposed approach was tested using different number of training images per class. The used features for
classification is shape feature extraction based on edge detection and wavelet transform. Moreover, SVMs algorithm
was employed with different kernel functions that are: Linear kernel, radial basis function (RBF) kernel, and MultiLayer
Perceptron (MLP) kernel for for water quality degree classification.
Figure 4 depicts experimental results that show classification accuracy obtained via applying each kernel function
considering 10, 20, 30, and 35 training images per class. The results of the proposed classification approach were
evaluated against human expert assessment for measuring obtained accuracy. As shown in figure 4, with the linear
kernel function being used for SVMs algorithm and the number of training images per class is 35, the proposed
classification approach achieved 95.41 Ψ accuracy for all water quality degrees. The accuracy is computed using
equation (12).
ݏ݁݃ܽ݉݅ ݂݀݁݅݅ݏݏ݈ܽܿ ݕ݈ݐܿ݁ݎݎ݋͓ܿ ൌ ݕܿܽݎݑܿܿܽ
ͳͲͲሺͳʹሻ כ ݏ݁݃ܽ݉݅ ݃݊݅ݐݏ݁ݐ͓
Many points of research assessing water pollution based on using fish gill microscopic images as biomarker
but this research using experimental laboratory as in paper [10]. However, none of them used not computer-based
system on the experimented dataset(s). So, to the best of our knowledge, this article is the first research work aims at
highlighting the most appropriate classification algorithm, for classifying water quality degree using fish gills
Figure 4: SVMs classification with different kernel functions &10-fold cross validation accuracy
results
610 Asmaa Hashem Sweidan et al. / Procedia Computer Science 65 ( 2015 ) 601 – 611
microscopic images as biomarker for assessing water pollution based on the shape features.
6. Conclusions and Future Work
In these studies, many feature extraction and classification approaches were used. In this study is designed
for fulfilling the system and testing the performance of the proposed Method for classification of fish gills in
microscopic images. The proposed approach consists of three main phases; pre-processing, feature extraction and
classification phase. edge detection and wavelet transform as shape feature vectors are combined to extract feature
and used as a PCA inputs for transformation. Finally, SVMs with 10 cross validation model is developed for water
quality degree classification. Based on the obtained results, water quality degree classification accuracy is 95.41%
using SVMs linear kernel function.
This research is intended to highlight the capabilities of usage of image classification technique based on fish
gills as biomarker for water quality. We aim to improve the understanding of fish gills as biomarker and effects of
fish on water quality.
REFERENCES
1. Reddy P, Rawat S. S, Assessment of Aquatic Pollution Using Histopathology in Fish as a Protocol,
International Research Journal of Environment Sciences, vol. 2, no. 8, pp. 79-82, 2013.
2. Rachel Ann Hauser-Davis, Reinaldo Calixto de Campos, and Roberta Louren Ziolli Fish Metalloproteins as
Biomarkers of Environmental Contamination, Springer., vol. 218, pp. 101–123, 2012.
3. Wael A. Omar, Khalid H. Zaghloul, Amr A. Abdel-Khalek, and S. Abo-Hegab, Risk Assessment and Toxic
Effects of Metal Pollution in Two Cultured and Wild Fish Species from Highly Degraded Aquatic Habitats,
Archives of environmental contamination and toxicology , vol. 65, no. 4, pp. 753-764,2013.
4. Abdolreza Jahanbakhshi and Aliakbar Hedayati, Gill histopathological changes in Great sturgeon after
exposure to crude and water soluble fraction of diesel oil, Comparative Clinical Pathology, vol. 22, no. 6, pp.
1083–1086,2013.
5. Morina Valon,Aliko Valbona, Eldores Sula, Gavazaj Fahri,Kastrati Dhurata and Cakaj Fatmir,
Histopathologic Biomarker of Fish Liver as Good Bioindicator of Water Pollution in Sitnica River, Kosovol,
Global Journal Of Science Frontier Research , vol. 13 , no. 5,2013.
6. Iman M.K. Abumourad, Mohammad M.N. Authman and Wafaa T. Abbas, Heavy Metal Pollution and
Metallothionein Expression: A Survey on Egyptian Tilapia Farmsl, Journal of Applied Sciences Research, vol. 9,
no. 1, pp. 612-619,2013.
7. Freylan Mena Torres, Maria Azzopardi, Sascha Pfenn

ผลลัพธ์ (ไทย) 1:

First and second masks are obtained by placing two different thresholds on the horizontal , vertical, anddiagonal components. Two different thresholds are used in this approach to get more edge information.x Third mask is obtained by looking at the maximum pixel value of the horizontal , vertical, and diagonalcomponents.x Forth mask is obtained by finding the max intensity pixels among h, v and d components and by multiplyingwith approximation component.Steps of Edge detection with wavelet that show the details of the feature extraction are describes as follows: Asmaa Hashem Sweidan et al. / Procedia Computer Science 65 ( 2015 ) 601 – 611 6071. After pre-processing phase Decompose the image using wavelets.2. Get approximation a, horizontal y, vertical v and diagonal d components.3. Filter out the strong edges of horizontal, vertical and diagonal components by using T1ߛ and T2ߛwhere T is threshold and ߛ is the standard deviation of respected image.4. Obtain first edge map by applying T1ߛ on h, v and d components and combining them and thenmultiplying the resulting mask with the approximation component.5. Repeat step5 on T2ߛ6. Repeat step5 on T1ߛ and T2ߛ7. Repeat step5 on finding the max intensity pixels among h, v and d components and by multiplyingwith approximation componentIn this paper moment invariants are used to represent a shape . Feature vectors are calculated for input imageand image data base. shape feature vector are given by equations (5-11).߶ሺͳሻ ൌ ߟଶ଴ ൅ ߟ଴ଶ (5)߶ሺʹሻ ൌ ሺߟଶ଴ െ ߟ଴ଶሻଶ ൅ Ͷߟଵଵଶ (6)߶ሺ͵ሻ ൌ ሺߟଷ଴ െ ͵ߟଵଶሻଶ ൅ ሺ͵ߟଶଵ െ ߟ଴ଷሻଶ (7)߶ሺͶሻ ൌ ሺߟଷ଴ ൅ ߟଵଶሻଶ ൅ ሺߟଶଵ ൅ ߟ଴ଷሻଶ (8)߶ሺͷሻ ൌ ሺߟଷ଴ െ ͵ߟଵଶሻሺߟଷ଴ ൅ ߟଵଶሻሾሺߟଷ଴ ൅ ߟଵଶሻଶ െ ͵ሺߟଶଵ ൅ ߟ଴ଷሻଶሿ ൅ ሺ͵ߟଶଵ െ ߟ଴ଷሻሺߟଶଵ ൅ ߟ଴ଷሻሾ͵ሺߟଷ଴ ൅ߟଵଶሻଶ െ ሺߟଶଵ ൅ ߟ଴ଷሻଶሿሺͻሻ߶ሺ͸ሻ ൌ ሺߟଶ଴ െ ߟ଴ଶሻሾሺߟଷ଴ ൅ ߟଵଶሻଶ െ ሺߟଶଵ ൅ ߟ଴ଷሻଶሿ ൅ Ͷߟଵଵሺߟଷ଴ ൅ ߟଵଶሻሺߟଶଵ ൅ߟ଴ଷሻሺͳͲሻ߶ሺ͹ሻ ൌ ሺ͵ߟଶଵ െ ߟ଴ଷሻሺߟଷ଴ ൅ ߟଵଶሻሾሺߟଷ଴ ൅ ߟଵଶሻଶ െ ͵ሺߟଶଵ ൅ ߟ଴ଷሻଶሿ െ ሺߟଷ଴ െ ͵ߟଵଶሻሺߟଶଵ ൅ ߟ଴ଷሻሾ͵ሺߟଷ଴൅ߟଵଶሻଶ െ ሺߟଶଵ ൅ ߟ଴ଷሻଶሿሺͳͳሻThe PCA algorithm has been utilized for feature extraction. PCA transformed space contribution to the sub-spacesto reduce the dimensions. Then, it turns the contribution of space in sub-spaces to reduce the dimensions. A shapefeature vector will be formed as a four edge maps.4.3. Classification phaseIn this stage, SVMs classifier is used for classification of feature vectors from features extracted stage viaclassifying input image using a trained model. This phase employs retrieving all the best matching images from thematching class of the input image, whereas the outputs are the corresponding water quality degree equivalent to eachimage in the testing dataset. The proposed approach addresses a multi-class problem where a variety of techniqueswas used for decomposition of the multi-class problem into several binary problems using SVMs as binary classifiers.In this research, We used one-against-all approach with 10-fold cross validation to solve a multi-class problems.Algorithm (1) shows the details about the classification algorithm.5. Experimental Analysis and DiscussionNile Tilipia "Oreochromis niloticus” is pollution resistant species ideal for biomarker of water pollution. Thedatasets used for experiments were constructed based on real sample microscopic images for fish gills in differenthistopathlogical change stages exposed to copper and water pH.Fish images were collected from Abbassa farm, Abo-Hammad, Sharkia Governote, Egypt. Some samples of bothtraining and testing datasets are shown in fig. 2. Training dataset is divided into 4 classes representing the different 608 Asmaa Hashem Sweidan et al. / Procedia Computer Science 65 ( 2015 ) 601 – 611water quality degrees; namely excellent quality, good quality, moderate quality, and bad quality [19], as shown in fig.3เลือกอัลกอริทึม (1) ลักษณะการทำงานและฝึกอบรม classifierFig.2 ตัวอย่างของการฝึกอบรม และทดสอบปลา gills ภาพด้วยกล้องจุลทรรศน์x คุณภาพดีน้ำ: แสดงหลักใย (F) และรอง lamellae (L) เกิดจากเหล่านี้3.a ขนานกับแต่ละอื่น ๆ และตั้งฉากกับแกนใย เป็นแสดงในรูปx คุณภาพดีน้ำ: ที่ปลาสัมผัสกับทองแดงที่ pH 9 ได้มากหรือน้อยคล้ายกับผู้ควบคุมกลุ่ม ดังที่แสดงในรูปที่ 3.bคุณภาพน้ำปานกลาง x: ที่ปลาสัมผัสกับทองแดงที่ pH การเจริญเกินแสดง 7 ของ lamellae หลักมีอาหารที่สมบูรณ์ของ lamellaeand รองตัดให้สั้นลง ดังที่แสดงในรูปที่ 3.cx คุณภาพน้ำไม่ดี: ที่ปลาสัมผัสกับทองแดงที่ pH 5 Drooping และทำให้สั้นของบางlamellae รอง เป็นแสดงในรูป 3.dป้อนข้อมูล: ข้อมูลการฝึกอบรมเป็นอย่างܺ଴ൌሾݔͳǡݔʹǡǤǤǤݔ݈ሿ்เริ่มต้น: ดัชนีสำหรับคุณลักษณะที่เลือก: f = [1, 2,... n]รถไฟ classifier SVMs ใช้ܺ଴ตัวอย่างสำหรับ t = 1,..., โด Tคำนวณเกณฑ์การจัดอันดับตาม SVMs ฝึกเลือก ܯ௧features ด้านบน และกำจัดคุณลักษณะอื่น ๆตัวอย่างการฝึกอบรมเพื่อเลือกคุณลักษณะจำกัดสร้างܰ SVMs ไบนารีSVM แต่ละแยกชั้นจากชั้นอื่น ๆรถไฟ݅௧௛ SVM มีตัวอย่างการฝึกอบรมทั้งหมดของ݅ระดับ௧௛บวกป้ายชื่อ และตัวอย่างการฝึกของชั้นเรียนอื่น ๆ ด้วยป้ายชื่อที่ติดลบสิ้นสุดสำหรับแสดงผล: จำแนกระดับคุณภาพน้ำ Asmaa Hashem Sweidan et al. / Procedia Computer Science 65 ( 2015 ) 601 – 611 609Fig. 3. Examples of water quality degreesThe proposed approach was tested using different number of training images per class. The used features forclassification is shape feature extraction based on edge detection and wavelet transform. Moreover, SVMs algorithmwas employed with different kernel functions that are: Linear kernel, radial basis function (RBF) kernel, and MultiLayerPerceptron (MLP) kernel for for water quality degree classification.Figure 4 depicts experimental results that show classification accuracy obtained via applying each kernel functionconsidering 10, 20, 30, and 35 training images per class. The results of the proposed classification approach wereevaluated against human expert assessment for measuring obtained accuracy. As shown in figure 4, with the linearkernel function being used for SVMs algorithm and the number of training images per class is 35, the proposedclassification approach achieved 95.41 Ψ accuracy for all water quality degrees. The accuracy is computed usingequation (12).ݏ݁݃ܽ݉݅ ݂݀݁݅݅ݏݏ݈ܽܿ ݕ݈ݐܿ݁ݎݎ݋͓ܿ ൌ ݕܿܽݎݑܿܿܽͳͲͲሺͳʹሻ כ ݏ݁݃ܽ݉݅ ݃݊݅ݐݏ݁ݐ͓Many points of research assessing water pollution based on using fish gill microscopic images as biomarkerbut this research using experimental laboratory as in paper [10]. However, none of them used not computer-basedsystem on the experimented dataset(s). So, to the best of our knowledge, this article is the first research work aims at
highlighting the most appropriate classification algorithm, for classifying water quality degree using fish gills
Figure 4: SVMs classification with different kernel functions &10-fold cross validation accuracy
results
610 Asmaa Hashem Sweidan et al. / Procedia Computer Science 65 ( 2015 ) 601 – 611
microscopic images as biomarker for assessing water pollution based on the shape features.
6. Conclusions and Future Work
In these studies, many feature extraction and classification approaches were used. In this study is designed
for fulfilling the system and testing the performance of the proposed Method for classification of fish gills in
microscopic images. The proposed approach consists of three main phases; pre-processing, feature extraction and
classification phase. edge detection and wavelet transform as shape feature vectors are combined to extract feature
and used as a PCA inputs for transformation. Finally, SVMs with 10 cross validation model is developed for water
quality degree classification. Based on the obtained results, water quality degree classification accuracy is 95.41%
using SVMs linear kernel function.
This research is intended to highlight the capabilities of usage of image classification technique based on fish
gills as biomarker for water quality. We aim to improve the understanding of fish gills as biomarker and effects of
fish on water quality.
REFERENCES
1. Reddy P, Rawat S. S, Assessment of Aquatic Pollution Using Histopathology in Fish as a Protocol,
International Research Journal of Environment Sciences, vol. 2, no. 8, pp. 79-82, 2013.
2. Rachel Ann Hauser-Davis, Reinaldo Calixto de Campos, and Roberta Louren Ziolli Fish Metalloproteins as
Biomarkers of Environmental Contamination, Springer., vol. 218, pp. 101–123, 2012.
3. Wael A. Omar, Khalid H. Zaghloul, Amr A. Abdel-Khalek, and S. Abo-Hegab, Risk Assessment and Toxic
Effects of Metal Pollution in Two Cultured and Wild Fish Species from Highly Degraded Aquatic Habitats,
Archives of environmental contamination and toxicology , vol. 65, no. 4, pp. 753-764,2013.
4. Abdolreza Jahanbakhshi and Aliakbar Hedayati, Gill histopathological changes in Great sturgeon after
exposure to crude and water soluble fraction of diesel oil, Comparative Clinical Pathology, vol. 22, no. 6, pp.
1083–1086,2013.
5. Morina Valon,Aliko Valbona, Eldores Sula, Gavazaj Fahri,Kastrati Dhurata and Cakaj Fatmir,
Histopathologic Biomarker of Fish Liver as Good Bioindicator of Water Pollution in Sitnica River, Kosovol,
Global Journal Of Science Frontier Research , vol. 13 , no. 5,2013.
6. Iman M.K. Abumourad, Mohammad M.N. Authman and Wafaa T. Abbas, Heavy Metal Pollution and
Metallothionein Expression: A Survey on Egyptian Tilapia Farmsl, Journal of Applied Sciences Research, vol. 9,
no. 1, pp. 612-619,2013.
7. Freylan Mena Torres, Maria Azzopardi, Sascha Pfenn

ผลลัพธ์ (ไทย) 3:

หน้ากากตัวแรกและตัวที่สองจะได้รับโดยการวางสองเกณฑ์ที่แตกต่างกันในแนวนอน แนวตั้ง และแนวทแยง
ส่วนประกอบ . สองเกณฑ์ที่แตกต่างกันจะใช้ในวิธีการนี้จะได้รับเพิ่มเติมขอบข้อมูล .
x 3 รูปแบบได้โดยดูที่ค่าพิกเซลสูงสุดของแนวนอน แนวตั้ง และแนวทแยงส่วนประกอบ
.
x ออกมาเป็นหน้ากากได้โดยการหาสูงสุดความหนาแน่นพิกเซลของเอชวี d ส่วนประกอบและส่วนประกอบโดยการคูณด้วย

ประมาณ ขั้นตอนของการตรวจหาขอบด้วยเทคนิคที่แสดงรายละเอียดของลักษณะการสกัด มีอธิบายดังนี้ :
asmaa Hashem sweidan et al . วิทยาศาสตร์คอมพิวเตอร์ / procedia 65 ( 2015 ) 601 – 611 607
1 หลังจากการประมวลผลเฟสเน่าภาพใช้ wavelets .
2 ได้ประมาณเป็น แนวนอน คแนวตั้งและแนวทแยง D ส่วนประกอบ .
3 กรองออกจากขอบที่แข็งแกร่งของแนวนอน แนวตั้ง และแนวทแยงส่วนประกอบโดยใช้ߛ T1 และ T2 ที่ߛ
t และระดับߛเป็นส่วนเบี่ยงเบนมาตรฐานของภาพที่เคารพ .
4 ขอรับแผนที่ขอบแรกโดยการใช้ߛ T1 h , V และ D ส่วนประกอบและการรวมแล้ว
คูณเป็นผลหน้ากากด้วยการประมาณส่วนประกอบ .
5 ย้ำ step5 ใน T2 ߛ