International Conference for Convergence of Technology - 2014

Unhealthy Region of Citrus Leaf Detection Using

Image Processing Techniques
Ms. Kiran R. Gavhale Prof. Ujwalla Gawande Mr. Kamal O. Hajari
Dept. of Computer Technology Asst. Prof. Dept. of Computer Dept. of Information Technology
Yeshwantrao Chavan College of Technology Yeshwantrao Chavan College of
Engineering Yeshwantrao Chavan College of Engineering
Nagpur, Maharashtra, India Engineering Nagpur, Maharashtra, India
E-mail: kiran89gavhale@gmail.com Nagpur, Maharashtra, India E-mail: kamalhajari123@gmail.com
E-mail: ujwallgawande@yahoo.co.in

Abstract— Producing agricultural products are difficult task Common symptoms are includes abnormal leaf growth,
as the plant comes to an attack from various micro-organisms, color distortion, stunted growth, shriveled and damaged
pests and bacterial diseases. The symptoms of the attacks are pods. Although diseases and insect pests can cause
generally distinguished through the leaves, steams or fruit considerable yield losses or bring death to plants and it’s
inspection. The present paper discusses the image processing also directly affect to human health. These require careful
techniques used in performing early detection of plant diseases
diagnosis and timely handling to protect the crops from
through leaf features inspection. The objective of this work is
to implement image analysis and classification techniques for heavy loses [2]. In plant, diseases can be found in various
extraction and classification of leaf diseases. Leaf image is parts such as fruit, stem and leaves. Leaf presents several
captured and then processed to determine the status of each advantages over flowers and fruits at all seasons worldwide
plant. Proposed framework is model into four parts image pre- [3, 4].
processing including RGB to different color space conversion, This enables machine vision to provide image based
image enhancement; segment the region of interest using K- automatic detection and guidance. This paper is organized
mean clustering for statistical usage to determine the defect into the following sections. Section 1 gives an introductory
and severity areas of plant leaves, feature extraction and part and importance of leaf disease detection. Section 2
classification. texture feature extraction using statistical
describes various types of leaf diseases and its symptoms.
GLCM and color feature by means of mean values. Finally
classification achieved using SVM. This technique will ensure Section 3 presents a detailed discussion on recent work
that chemicals only applied when plant leaves are detected to carried out in this area. Section 4 includes proposed
be effected with the disease. methodology for leaves disease extraction and classification
which represents a brief review on various image processing
Keywords: citrus; canker; anthracnose; co-occurrence techniques. Section 5 shows experimental result. Finally,
matrix; SVM; texture feature. Section 6 concludes this paper along with possible future
directions.
I. INTRODUCTION
II. PLANT DISEASES ANALYSIS AND ITS SYMPTOMS
Agriculture is the mother of all cultures. It has played an
important role in the development of human civilization. Proposed work focusing on citrus trees which include
The focus on enhancing the productivity, without grapefruit, lemons, limes and oranges leaf attack by various
considering ecological impacts has resulted into citrus diseases. Some citrus leaves diseases and it’s
environmental degradation. Without any adverse symptoms as shown below:
consequences, improvement of the productivity can be done
in a sustainable manner. Plants exist everywhere we live, as A. Citrus canker
well as places without us. Lots of them carry significant Citrus canker is spread by wind driven rain. Typical citrus
information for the development of human society. As canker lesions on plant leave having range from 2-10 mm in
diseases of the plants are inevitable, detecting disease plays size and seem as concentric circles on the underneath of the
a major role in the field of Agriculture [1]. Plant leaf disease leaf. Often lesions will be encircled by a water-soaked
is one of the crucial causes that reduces quantity and margin and yellow color halo. As canker lesion matures, it
degrades quality of the agricultural products. Currently may lose it palpable roughness but concentric circles still
chemicals are applied to the plants periodically without visible. The yellow halo eventually changes to dark brown
knowing the requirement of plants. or black color. Water-soaked edge adjoining the lesion may

978-1-4799-3759-2/14/$31.00©2014 IEEE 1
 International Conference for Convergence of Technology - 2014

reduce as shown in fig.1 (a). The middle of the lesion will diseases detection (HPCCDD) algorithm has been used for
be rough in texture with a pimple-like point. Always found categorizing the diseases. They claim the accuracy of 98.1%
in both sides of the leaf with the exception of very young over existing algorithm. Tushar H Jaware et al. [4]
lesions. proposed a novel , improved k-means clustering technique
to solve low-level image segmentation. Spatial gray-level
B. Anthracnose
dependence matrices (SGDM) method has been used for
Anthracnose is a key colonizer of injured and senescent extracting statistical texture features by Sanjay B. Dhaygude
tissue. The micro-organism grows on lifeless wood in the et al. [5] also RGB images have been converted into Hue
canopy, and then spreads short distances by heavy dew, rain Saturation Value (HSV) color space representation and
splash, and overhead irrigation. Common symptoms are a showed the H, S and V components. Mokhled S. Al-
more or fewer circular, flatted area, light tan in color with a Tarawneh [6] presented an empirical investigation of olive
prominent purple margin as shown in figure 1(b). leaf spot disease using auto-cropping segmentation and
C. Overwatering fuzzy c-means classification. RGB to LAB color space and
median filter used for image enhancement. At end present
IT is another cause of unhealthy leaves due to
comparative assessment of fuzzy c-means and k-mean
overwatering. Large amount of water can cause curl leaves, clustering.
turn yellow color and drop as shown in fig.1(c). S.M.Ramesh, et.al.[7] proposed new technique for
D. Citrus greening disease enhancement of color images by scaling the discrete cosine
Leaves with yellow veins formation is the most transform coefficients. Provides better enhancement
characteristic symptom of citrus greening disease. Some compared to image capture by digital camera if resolution
times look like leaf deficiency as shown in fig.1 (d). improper. Back-propagation (BP) networks have been used
to classify the grape and wheat diseases by Haiguang Wang
et al. [8]. Also principal component analysis (PCA) has
been used to reduce dimensions of the feature data. A.
Menukaewjinda et al. [9] tried another ANN, i.e. back
propagation neural network (BPNN) for efficient grape leaf
color extraction with complex background. They also
explore modified self organizing feature map (MSOFM) and
(a) Canker (b) Anthracnose genetic algorithm (GA) and found that these techniques
provide automatic adjustment in parameters for grape leaf
disease color extraction. Support vector machine (SVM) has
been also found to be very promising to achieve efficient
classification of leaf diseases. Detection of unhealthy region
and classification using texture features has been proposed
(c) Overwatering (d) Citrus Greening by S. Arivazhagan, et al. [10]. Their algorithm has been
tested on ten species of plants namely banana, beans,
Fig. 1. Citrus leaves diseases
jackfruit, lemon, mango, potato, tomato and sapota. 94.74%
accuracy has been achieved by Support vector machine
III. LITERATURE REVIEW (SVM) classifier.
A Research of maize disease image recognition of corn
A Proliferation of literature is available in plant leaf based on BP networks effectively identified by Song Kai et
disease detection. We will highlight some of the key al. [11] where YCbCr color space technology is used to
contributions. segment disease spot, Co-occurrence matrix (CCM) spatial
A methodology for detecting plant diseases early and gray level layer is used to extract disease spot texture
accurately using diverse image processing techniques has feature, BP neural network has been used to classify the
been proposed by Anand H.Kulkarni et al. [1], in which maize disease. The applications of K-means clustering as
Gabor filter has been used for feature extraction and ANN well as BP neural networks had been formulated for
based classifier has been used for classification with clustering and classification of diseases that affect on plant
recognition rate up to 91%. F. Argenti, et al. [2] proposed a leaves by H. Al-Hiary, et al. [12]. They provide adequate
fast algorithm for calculating parameters of co-occurrence support for accurate detection of leaf diseases.
matrix by supervised learning and maximum likelihood IV. PROPOSED METHODOLOGY
method for fast classification. Homogenize techniques like
canny and sobel filter has been used to identify the edges by
There are five main steps used for the detection of plant
P.Revathi et al. [3]. Then these extracted edge features have
leaf diseases as shown in fig.2. The processing scheme
been used in classification to identify the disease spots.
consists of image acquisition through digital camera or
Proposed homogeneous pixel counting technique for cotton

978-1-4799-3759-2/14/$31.00©2014 IEEE 2
 International Conference for Convergence of Technology - 2014

scanner, image pre-processing includes image enhancement, coefficients, preservation of color by Scaling DC and AC
image segmentation where the affected and useful area are coefficients using function including all three colors
segmented, feature extraction and classification. Lastly the information [7].
presence of diseases on the plant leaf will be identified.
Here we present step by step approach for segment the
diseased image and extract its features.
A. Image Acquisition
Firstly, the images of various leaves acquired using a
digital camera with required resolution for better quality or
by scanner. The input photo image is then resized to
256x256 pixels. The construction of an image database is a) Original image b) CES Ehnhanced
Fig.3. Image Enhancement using CES
clearly dependent on the application. The image database
itself is responsible for the better efficiency of the classifier
which decides the robustness of the algorithm. 2) Colorspace conversion: This block play important
role in leaf spot detection because it defines the needed
B. Image Pre-processing color transformation that will be used in cluster separation.
Image pre-processing is the lowest level of abstraction The RGB images of citrus leaf are converted into color
whose aim is to improve the image data that suppress space representation. The principle of color space is to
undesired distortions as well as enhances some image facilitate the specification of colors in some standard,
features which is important for further processing and generally accepted way.
analysis task. It includes color space conversion and image There are various color space available such as CIELAB,
enhancement. Ycbcr, HSV. Color transformation determines the
luminosity and chromaticity layers which classify colors
according to human vision system and that very useful in
I. Image Acquisition II. Image
subjective analysis of leaf images judgment. RGB is an
Preprocessing
1. Pick up a leaf from plants ideal for color generation while HSV model is an ideal tool
Image Enhancement for color perception [5]. Hue is a color attribute that
using SF-CES describes pure color as perceived by an observer. Saturation
refers to relative purity or the amount of white light added to
2. Capture digital leaf image Color-space hue. Value means amplitude of light.
conversion

IV. Feature Extraction III. Image

using GLCM Segmentation using
K-mean clustering

V. Classification

Fig. 2. Proposed Methodology

1) Image enhancement: Direct observation of color

images is often strikingly different as human visual
perception computes the conscious representation. A simple
approach in the block (DCT) Discrete Cosine Transform
domain for enhancing color images by scaling the transform
coefficients i.e. color enhancement by scaling (SF-CES) .
The unique feature is that it uses chromatic and luminance
component which improves the visual quality of the images
to a great extent. It provides better enhancement for color
Figure 4: An example of K-Means clustering output for a citrus leaf that is
images as shown in fig.3. There are also other advantages infected with anthracnose disease. (a) Original image , (b) Enhanced image,
for using this algorithm such as adjustment of local (c) Ycbcr color space conversion, (d, e, f) pixel of the first cluster shows
background brightness by scaling DC coefficients, green part of leaf while second and third clusters shows disease part
respectively.
preservation of local contrast by scaling DC and AC

978-1-4799-3759-2/14/$31.00©2014 IEEE 3
 International Conference for Convergence of Technology - 2014

a) YCbCr : YCbCr color system is a common color achieved by minimizes the sum of squares of distances of
system used here which is applied by the most widely used the objects and the corresponding cluster. However, K-
JPEG image as well as for digital video. In this color space means clustering is used to separate the leaf image into
system, luminance information is stored as a single different clusters if a leaf contains more than one disease.
luminance component ‘Y’, and chrominance information is The suitable color group numbers lead to the better color
stored in two color-difference components i.e. Cb and Cr. extraction [9]. An example of K-Means clustering output for
Where Cb represents the difference between blue a leaf infected with citrus anthracnose disease shown in fig.
component and reference value. Cr represents the difference 4, 5, and 6 with different color space.
between red component and a reference value. It is used in It is observed from figure 5 and 6 where cluster 1 and 2
bandwidth reduction, sub sampled, and also treated contains infected object of canker anthracnose disease and
separately for improved system efficiency [12]. RGB color clusters 3 contain green part of leaf using Ycbcr and Lab
transformation into YCbCr implemented and the result as color space.
shown in fig.4.
D. Feature Extraction
b) L*a*b* : L*a*b* color space chosen due to the
uniform distribution of colors as it visualize same as human After segmentation the area of interest i.e. diseased part
perception of color [6]. It consists of a luminosity layer 'L*', extracted. The aim of this phase is to find and extract
chromaticity layer 'a*' represents red-green axis and features that can be used to determine the meaning of a
chromaticity-layer 'b*' represents where the color place given sample. Image features usually include color, shape
along the blue-yellow axis. RGB color transformation into and texture features in image processing [2]. Texture is one
L*a*b* implemented and the result as shown in fig. 5. of most popular features for image classification and
retrieval. Currently most of the researchers targeting plant
leaf texture as the most important feature in classifying
plants.
In this part, we present a statistical method using GLCM.
This method allows computing some statistics describing
texture. In this section, different statistics used to describe
texture after construction of the GLCMs present.
Gray-Level Co-Occurrence Matrix (GLCM) :GLCM is
the statistical method of investigative texture which
considers the spatial relationship of pixels [10]. The GLCM
functions characterize the texture of a leaf image by
calculating occurrence of pixel in an image with specific
values and in a specified spatial relationship. By creating a
GLCM followed by extracting statistical measures from this
matrix.
Gray-Level Co-Occurrence matrix creation :
Graycomatrix function creates a gray-level co-occurrence
matrix (GLCM) by calculating occurrence of a pixel value i
with the pixel value j in a specific spatial relationship.
Figure 5: An example of K-Means clustering output for a citrus leaf that is
infected with anthracnose disease. (a) Original image , (b) Enhanced image, Specifying the Offsets: To create number of GLCMs, have
(c) LAB color space conversion, (d, e, f) pixel of the first cluster shows to specify an array of offsets. Offsets describe pixel
green part of leaf while second and third clusters shows disease part relationships of varying direction and distance such as four
respectively.
directions viz. horizontal, vertical, two diagonals and four
distances. Represent these offsets as integer’s p-by-2 array.
C. Image Segmentation Each row in an array is a two-element vector i.e.
Image segmentation is process used to simplify the [row_offset, col_offset].
representation of an image into something that is more Derive Statistics from a GLCM: Here, statistics derived
meaningful object of interest from background and easier to from GLCMs using the graycoprops function. Hence, we
analyze. Segmentation is also done through feature based will see statastics viz. contrast, energy, homogeneity and
clustering [8]. Basic clustering k-means algorithm is used entropy [5, 11].
for segmentation in textured images [6].Creating device- 1) Contrast: Contrast measures intensity contrast
independent color space conversion as discussed in previous of a pixel and its neighbor pixel over the entire image. If the
phase in which coordinates used to specify the color. image is constant, contrast is equal to 0 while the largest
The K-means clustering algorithm used to classify pixels value can be obtained when the image is a random intensity
based on a set of features [4, 12]. The classification

978-1-4799-3759-2/14/$31.00©2014 IEEE 4
 International Conference for Convergence of Technology - 2014

image while that neighbor intensity and pixel intensity are training data point. In general larger the margin lowers
very different. Equation of the contrast is as follows: the generalization error of the classifier. The feature vector
is given as input to the classifier. This method differentiates
and identifies the citrus canker and citrus anthracnose.
Classification and recognition randomly divide database for
training and testing. Both subsets have the random samples
2) Energy: Energy is a measure of uniformity with from the same distribution. Training dataset consist of 200
squared elements summation in the GLCM. Range is in sample images of both citrus leave i.e. canker and
between 0 and 1. Energy is 1 for a constant image. The
anthracnose and testing dataset consist of 50 images of each
equation of the energy is shown: disease sample. The classifier train on the training set
applies it to the testing set and then measure performance by
comparing the predicted labels and give decision as canker
and anthracnose as shown in fig.7.

3) Homogeneity: Homogeneity measures the

closeness of the distribution of elements towards the
diagonal in the GLCM. It’s range in between 0 and 1.
Homogeneity is 1 for a diagonal GLCM. The equation of
the contrast is as follows:

Fig.7. Classifier types

4) Correlation: Correlation measures how correlat a
pixel to its neighbor pixel over the whole image. Range is in V.EXPERIMENTAL RESULTS
between -1 and 1.
TABLE I. TRAINING AND TESTING SET
No. of images used No. of images used
Classifiers Total
Where , for training for testing
Pij = Element i,j of the image SVMRBF
200 100 300
N = Number of gray levels SVMPOLY

μ = mean value of all pixels in the relationships that

contributed to the GLCM, it is calculated as: TABLE II. FAR, FRR AND GAR RATE
Classifiers FAR% FRR% GAR%
SVMRBF 4% 3% 96%
Б 2 = variance of the intensities of all reference pixels SVMPOLY 5% 6% 95%
calculated as:
The reference citrus leaf images database is created and
on that database different classification techniques are
applied such as support vector machine (SVM) with radial
basis kernel and polynomial kernel and several experimental
RGB representation of image is used for color textures results are obtained. The different training and testing
features extraction. Features vector (FV) is computed by samples are used while performing training and testing
features extraction (FE) for every R (Red), G (Green) and B using classifier mention in table-II and the accuracy of the
(Blue) color channel. system is based on the performance evaluation parameter
E. Classifier i.e. false acceptance rate (FAR) is 4%, false recognition rate
(FRR) is 3% and the genuine acceptance rate (GAR) is 96%
The Support Vector Machine (SVM) classifiers for the SVMRBF classifier. The false acceptance rate (FAR)
differentiate citrus leaf disease i.e. canker and anthracnose. is 5%, false recognition rate (FRR) is 6% and the genuine
It constructs a hyper plane in a high dimensional space that acceptance rate (GAR) is 95% for the SVMPOLY classifier.
can be used for classification. Good separation is achieved
by the hyper plane that has the largest distance to the nearest

978-1-4799-3759-2/14/$31.00©2014 IEEE 5
 International Conference for Convergence of Technology - 2014

[8] Haiguang Wang, Guanlin Li, Zhanhong Ma, Xiaolong Li, “Image
Recognition of Plant Diseases Based on Back propagation Networks,
5th International Congress on Image and Signal Processing, pp-894-
900, Chongqing, China, 2012
[9] A.Menukaewjinda, P.Kumsawat, K.Attakitmongcol, A.Srikaew,
“Grape leaf disease detection from color imagery using hybrid
intelligent system”, Proceedings of electrical Engineering/electronics,
Computer, Telecommunications and Information technology (ECTI-
CON), vol 1. pp: 513-516, Krabi, Thailand, 2008 .
[10] S. Arivazhagan , R. Newlin Shebiah , S. Ananthi, S. Vishnu Varthini ,
“Detection of unhealthy region of plant leaves and classification of
plant leaf diseases using texture features”, Commission Internationale
du Genie Rural(CIGR) journal, vol. 15, no.1, pp:211-217, March
Fig. 8. Graph for the classification accuracy 2013.
[11] Song Kai, liu zhikun, Su hang, Guo chunhong , “A Research of maize
disease image recognition of Corn Based on BP Networks”, Third
VI. CONCLUSIONS International Conference on Measuring Technology and
Mechatronics Automation,pp:246-249, Shenyang , China, 2011.
[12] H. Al-Hiary, S. Bani-Ahmad, M. Reyalat, M. Braik and Z.
This paper presents a number of image processing ALRahamneh, “Fast and Accurate Detection and Classification of
techniques to extract diseased part of leaf. SF-CES provides Plant Diseases”, International Journal of Computer Applications,
better enhancement of color image, Lab and Ycbcr color Wageningen Academic publishers, vol. 17, no.1, pp: 31-38, March
space supports K-mean clustering for disease part extraction 2011.
through the means of clusters. Then GLCM texture feature
and color textures features are extracted for further
classification purpose. Finally classification based on SVM.
With this method, the use of harmful chemicals on
plants can be reduced and hence ensure a healthier
environment. We should make further research on test
equipment, image processing and analysis method to
achieve better detection accuracy with different plant
species.
REFERENCES
[1] Anand H. Kulkarni, Ashwin Patil R. K., “Applying image processing
technique to detect plant diseases”, International Journal of Modern
Engineering Research, vol.2, Issue.5, pp: 3661-3664, 2012.
[2] F. Argenti,L. Alparone,G. Benelli ,” Fast algorithms for texture
analysis using co-occurrence matrices” Radar and Signal Processing,
IEE Proceedings , vol. 137, Issue 6, pp:443-448 , No. 6, December
1990.
[3] P. Revathi, M. Hemalatha, “Classification of Cotton Leaf Spot
Diseases Using Image Processing Edge Detection Techniques”, IEEE
International Conference on Emerging Trends in Science,
Engineering and Technology, pp-169-173, Tiruchirappalli,
Tamilnadu, India, 2012.
[4] Tushar H. Jaware, Ravindra D. Badgujar and Prashant G. Patil, “Crop
disease detection using image segmentation”, National Conference on
Advances in Communication and Computing, World Journal of
Science and Technology, pp:190-194, Dhule, Maharashtra, India,
2012.
[5] Prof.Sanjay B. Dhaygude, Mr.Nitin P. Kumbhar, “Agricultural plant
Leaf Disease Detection Using Image Processing”, International
Journal of Advanced Research in Electrical, Electronics and
Instrumentation Engineering , S & S Publication vol. 2, Issue 1, pp:
599-602, 2013.
[6] Mokhled S. Al-Tarawneh “An Empirical Investigation of Olive Leave
Spot Disease Using Auto-Cropping Segmentation and Fuzzy C-
Means Classification”, World Applied Sciences Journal , vol.23, no.9,
pp:1207-1211,2013.
[7] S.M.Ramesh, Dr.A.Shanmugam, “A New Technique for
Enhancement of Color Images by Scaling the Discrete Cosine
Transform Coefficients”, International Journal of Electronics &
Communication Technology, IJECT Vol. 2, Issue 1, March 2011.

978-1-4799-3759-2/14/$31.00©2014 IEEE 6