PerLeaf
Image Classification Application for Plants Leaf and its 

diseases

Presented by: 
Tamer Aghbar

Supervised by:
Dr. Mohammad Sharaf
Dr. Anas Toma



Outline:
1. Introduction

1. Problem Statement
2. Plants Diseases
3. Image Acquisition

2. Image processing
1. Image cropping & resizing
2. Leaf Segmentation

3. Feature Extraction
1. Color Features
2. Texture Features
3. Shape Features
4. Model Training & Accuracies

4. Design & implementation
1. Mobile App
2. Server

5. Conclusion & Future work
6. Demo
7. Final speech
8. Discussion



Introduction: 

-  The science of cultivating plants, animals, and 
other life forms for food.

- The most important industry around the world.
- The base of the food pyramid
- The major industry in the United States.

Agriculture Industry

1



Introduction: 

Green Revolution

- Better Quality
- More Quantity 
- Better Income
- Better Life

2



Problem Statement:

3

 

[1] Purification and Phytotoxic Analysis of Botrytis cinerea Virulence 
Factors: New Avenues for Crop Protection



Plants Diseases:

4

1- Grape Black rot 2- Grape Leaf blight



Image Acquisition:

5



Dataset Samples:

Apple Blueberry Cherry Grape healthy

Grape Black rot Grape leaf blight Pepper Potato

tomatoStrawberryRaspberry Soybean6



Image Processing:

7.1



Image Processing:

7.2



Image Processing:

7.3



1- Image Cropping & Resizing

 

8



2- Leaf Segmentation:

?
9.1



2- Leaf Segmentation:

9.1



2- Leaf Segmentation:

9.2

• Shape Segmentation

• Texture Segmentation

• Color Segmentation



2- Leaf Segmentation:

9.3

• Color Segmentation:
  R  G  B
[47, 65, 23]



2- Leaf Segmentation:

 

If new_pixel_value > 0:
write(new_pixel_value, seg_img);

else
write(0, seg_img);

9.4



2- Leaf Segmentation:

9.5

Init_mask IC_mask IC_mask_or ICx_mask

ICx_mask_and

Final_mask_closing



2- Leaf Segmentation:

9.6



Feature Extraction

10.1

Objects Descriptors

How to describe our input?

It’s a Leaf
Grape Leaf
Not Healthy
Disease Name: Black rot



Feature Extraction

10.2

How to “Numeric” these “Descriptors?



Feature Extraction

10.3

Manual grouping

Random mess!!



Feature Extraction

10.4

Manual grouping

More Elegant!



Feature Extraction

10.5

Main Descriptors

1- Color Features

2- Texture Features

3- Shape Features



Feature Extraction

11.1

1- Color Features:

Green range:
 [(35, 0, 0), (75, 255, 255)]

Yellow range: 
[(25, 0, 0), (40, 255, 255)]

[LowerColorBoundary, UpperColorBoundary],



Feature Extraction

11.2

1- Color Features:



Feature Extraction

11.3

1- Color Features:

[ green_percentage, 
yellow_percentage, 
nonGreen_percentage, 
Mean, 
variance, 
standardDeviation, 
skewness]

= 7 color features

Final Output:



Feature Extraction

12.1

2- Texture Features:



Feature Extraction

12.2

2- Texture Features:
GLCM (Gray Level Co-occurrence Matrix)



Feature Extraction

12.3

2- Texture Features:

GLCM -> [0:12] Features

+

Correlation & Standard Deviation

=

[(0:12), correlation, standard_deviation] 15 Texture feature=

-> Haralick Texture Algorithm



Feature Extraction

13.1

3- Shape Features:



Feature Extraction

13.2

3- Shape Features:

3.1 – Eccentricity
E = W_max / H_max



Feature Extraction

13.3

3- Shape Features:

3.1 – Eccentricity
3.2 – Euler Number

• Euler = 1 - #H
• avg_areas
• avg_areas/leaf_area



Feature Extraction

13.4

3- Shape Features:

3.1 – Eccentricity
3.2 – Euler Number
3.3 – Signature



Feature Extraction

13.5

3.3 – Signature

3.3.1 – Signature Variance

 



Feature Extraction

13.6

3.3 – Signature

1st Derivative

3.3.2 – 1st Derivative

 



Feature Extraction

13.7

3.3 – Signature
1st Derivative

+ve

-ve

Zeros

3.3.2 – 1st Derivative



Feature Extraction

13.8

3.3 – Signature

3.3.2 – 1st Derivative

• # of corners = # of zero values (include the margin of threshold)
• Corners_percentage = # of corners / # of points
• Curvature = (∑nonZerosV alues) / (∑signatureV alues)



Feature Extraction

13.9

3- Shape Features:

3.1 – Eccentricity
3.2 – Euler Number
3.3 – Signature

AspectRation(Aspect), Compactness(C), Roundness(R), 
Roughness(G), Elongation(E), Solidity(S)

 

3.4 – other shape equations:



Feature Extraction

13.10

3- Shape Features:

[holes_number, holes_meanArea, cntHoles_Area, Eccentricity, 
Signature_variances_mean, Signature_corners_percent, 
Signature_Curvature, num_corners, Aspect, C, R, G, E, S] = 14 shape features

Final Output:



Feature Extraction

13.11



14.1

Model Training:

Random Forest Classifier



14.2

Accuracies:

• First Accuracy: ~56% (general shape features)
• Second Accuracy: ~83% (shape/color)
• Third Accuracy:

• Master Dataset: ~89.942% (Shape/Color/Texture)
• Grape Dataset: ~98.174% (Shape/Color/Texture)



14.3

Model Training & Accuracies

It’s a Leaf
Grape Leaf
Not Healthy
Disease: Black rot



15.1

Design & Implementation:



Design & Implementation:



15.2

Design & Implementation:



15.3

Design & Implementation:



15.4

Design & Implementation:



16.1

Conclusion & Future work:

1) Dataset

2) Features & Machine Learning

1) Build our own dataset
2) study more disease can be used in our model

1) Make more accurate features 
2) Compline similar features by doing some feature analysis,

In order to minimize the processing time



16.2

Conclusion & Future work:

3) Mobile Application:

1) Cross platform application
2) Build a website
3) Add dashboard screen
4) Users accounts for non-expert people and experts
5) Involve GPS & maps



Demo

17



Presented by: 
Tamer Aghbar

Supervised by:
Dr. Mohammad Sharaf

Dr. Anas Toma

Thank you!