International Journal of Computer Science and Telecommunications [Volume 3, Issue 6, June 2012] 65

Color Transform Based Approach for Disease Spot

Detection on Plant Leaf
ISSN 2047-3338

Piyush Chaudhary1, Anand K. Chaudhari2, Dr. A. N. Cheeran3 and Sharda Godara4

1,2,3
Electrical Department, Veermata Jijabai Technological Institute, Mumbai, India
4
Accenture Services Pvt. Ltd., Pune, Maharastra, India

Abstract— In this research, an algorithm for disease spot Nunik Noviana Kurniawati et al [3] introduced a method
segmentation using image processing techniques in plant leaf is for detection and classification of paddy disease. In this
implemented. This is the first and important phase for automatic method Otsu threshold is used for disease spot detection and
detection and classification of plant diseases. Disease spots are unnecessary spots are removed using median filter. Geng
different in color but not in intensity, in comparison with plant
Ying et al [5] studied the method of image pre-processing for
leaf color. So we color transform of RGB image can be used for
better segmentation of disease spots. In this paper a comparison detecting the disease spot. In this paper median filter is used
of the effect of CIELAB, HSI and YCbCr color space in the for image smoothing. Threshold technique is used to convert
process of disease spot detection is done. Median filter is used for filtered image into binary image and finally using edge
image smoothing. Finally threshold can be calculated by detection technique, disease spot is detected. Using above
applying Otsu method on color component to detect the disease techniques disease spot can be detected in “Monocot family”
spot. An algorithm which is independent of background noise, plants, in which mostly veins are parallel and less visible [9]
plant type and disease spot color was developed and experiments [10]. Problem occurs when the same technique is applied on
were carried out on different “Monocot” and “Dicot” family “Dicot family” plants to detect the disease spot, in which
plant leaves with both, noise free (white) and noisy background.
veins form a netted pattern. In dicot plant leaves, larger veins
Index Terms— CIELAB, Color Transform, Plant Leaf Spot are thicker and straighter [9], [10]. In the process of disease
Disease, Segmentation and Image Processing spot detection disturbance mainly occurs because of these
thicker veins.

I. INTRODUCTION

P LANT diseases are important factors, as it can cause

significant reduction in both quality and quantity of crops
in agriculture production. Therefore, detection and
classification of diseases is an important and urgent task.
Traditionally farmers identify the diseases by naked eye
observation method. In this method disease is visually
detected by the experts, who have the ability to detect subtle Monocot Family Plant Leaf [9] Dicot Family Plant Leaf [10]
changes in leaf color. This method is very laborious, time
consuming and impractical for large fields. Different experts Fig.1: Difference between Monocot and Dicot family plant Leaf
can detect same part as different disease. To increase accuracy
paper grid method is used. Drawback of this method is that Veins color is same as plant leaf color only intensity
this method is laborious. So a fast and accurate approach to differs. On the other hand, disease spot color is different from
identify the plant diseases is needed. plant leaf color. So first if image is transformed from device
Some researchers have used image processing techniques dependent color space to device independent color space and
for fast and accurate detection of plant diseases [1]-[8]. The threshold is applied on color component, one can get better
steps followed by these researchers in detection of leaf spot detection of disease spot. Some researchers first convert RGB
diseases are: image acquisition, image pre-processing, disease image into Hue Saturation Intensity (HSI) model and then
spot segmentation, feature extraction and disease apply threshold on H component for segmenting infected
classification. The accuracy of result depends on method used areas from plant leaves [1], [2]. Di Cui et al developed a fast
for disease spot detection. The main obstacle in disease spot manual threshold-setting method based on HSI color model to
detection is noise, which is introduced by camera flash, segment the disease spot [6]. Song Kai et al [4] convert RGB
change in illumination, noisy background and presence of image into YCbCr color space to detect the disease spot.
vein in the plant leaf. Therefore a method which wipes out the
noise and provides better disease spot segmentation is needed.

Journal Homepage: www.ijcst.org

 Piyush Chaudhary et al. 66

In this paper the effect of YCbCr, HSI and CIELAB color effect of presence of vein, RGB image should be color
space in the process of disease spot detection are compared. transformed before segmentation. After then Otsu threshold
Experiments were carried upon different “Monocot” and can be applied on color component to detect disease spot
“Dicot” family plant leaves with both noise free (white) and accurately. In this paper 3 color models are compared.
noisy background to get the method which is independent of
YCBCR Color Model
background noise and plant type.
This paper is organized in four sections. Methodology used This color model is widely used in digital video. In
is described in section 2 that includes 3 steps as follows, YCbCr color model, ‘Y’ indicates luminance component and
Image color transform, image smoothing and disease spot Cb, Cr indicates color component. Cb is the difference
segmentation. Experimentation and results are presented in between the blue components and CR is the difference
section 3. Conclusion to the paper is given in last section. between the red components [4] [13].
Using following formulas RGB image is transformed into
YCBCR color model [11].

 0.299   0.587  0.114  
II. METHODOLOGY
(1)
Fig. 2 shows the flowchart of the steps involved in the   0.168    0.331  0.500   (2)
disease spot detection. All the images in collection are in   0.500    0.418   0.081   (3)
JPEG format. These images are color transformed from RGB HSI (Hue Saturation Intensity) COLOR MODEL
image to one of the color space named by YCbCr, HIS and
CIELAB color spaces. The color transformed images are HSI is device dependent color model and based upon
passed through median filter to remove unnecessary spots. In Human Color Perception. In this color model ‘H’ indicates
last step Otsu threshold is applied on RGB image, ‘A’ Hue, which describes a pure color and is generally related to
component of CIELAB color space, ‘H’ component of HSI the wavelength of light. ‘S’ indicates Saturation, which
color space and ‘Cr’ component of YCbCr color space is used measures the “colorfulness” in HSI color Model. ‘I’ indicates
to detect the disease spot. The disease spot segmented images, Intensity, which shows the amplitude of the light [7][11][12].
    

obtained by all the three methods are compared to get the best
360     
(4)
method for disease spot detection.
With
$
&'(!)*+'(!,*-
  cos !" # %
$ .
&'(!)*% +'(!,*')!,*-%
Image Collection 6789'(,),,*
/01230145'/*  1  '(+)+,*
(5)
"
;51<5=1>';*  ' * (6)
6
Image Color Transform CIELAB Color Model
- YCbCr Color Model CIELAB system is device independent which is defined
- HIS Color Model by the CIE to classify color according to the human vision. In
- CIELAB Color Model the conversion process of an image from RGB color
component to CIELAB color component, first RGB image is
converted into CIEXYZ using following Equation [3].
?  0.4124   0.3576  0.1805   (7)
Image Smoothing

 0.2126   0.7152  0.722   (8)
@  0.0193   0.1192  0.9505   (9)

Disease Spot Segmentation Brightness and color information of LAB color model is
independent of each other. In CIELAB color model, ‘L’
describes color brightness; ‘A’ describes the color ranging
from green to red; ‘B’ describes the color ranging from blue
Comparison to yellow. Conversion Formula for LAB color model is [8].
A  0.2126   0.7152  0.0722   (10)
Fig. 2: Flow Chart for Disease Spot Detection using Image Processing
B  1.4749  '0.2213    0.3390  0.1177* 128 (11)
  0.6245  '0.1949   0.6057   0.8006  * 128 (12)
A. Image Color Transform
B. Image Smoothing
In plants, leaf vein is different in intensity and disease spot
is different in color, in comparison to plant leaf. So if Otsu During image collection, some noise may be introduced
threshold is applied on grayscale image, vein will also be because of camera flash. This noise can affect the detection of
present in binary image with the disease spot. But the region disease. To remove unnecessary spot, Image smoothing
of interest is only disease spots, not vein. For minimize the
International Journal of Computer Science and Telecommunications [Volume 3, Issue 6, June 2012] 67

technique is needed. In this paper median filter is used for this • Method 1: disease spots are segmented by applying Otsu
purpose. threshold on RGB image.
Median Filter • Method 2: in second method RGB image is first
converted into YCbCr color space using color transform
Median filter is a kind of higher order statistics filter.
formula. Then median filter is used for image
Median filter is nonlinear in nature, which replaces the value
smoothing. Disease spots are detected by applying Otsu
of the center pixel, by the median of the gray levels in the
threshold on ‘Cr’ component of filtered YCbCr color
image area enclosed by the filter. The median of a numerical
space.
collection is such that half the values in collection are less
than or equal to median, and half are greater than or equal to • Method 3: this is similar to method 2. Only difference is
median. that in place of YCbCr color space RGB image is
transformed into HSI color space and disease spots are
detected by applying Otsu threshold on ‘H’ component
of filtered HSI color space.
• Method 4: again same process is repeated using
CIELAB color space. Disease spots are segmented by
applying Otsu threshold on ‘A’ component of filtered
LAB color space.
As discussed in previous noise is introduced because of
camera flash, noisy background and veins in plant leaf. In
order to find the best method among these four methods,
research is categorized into three parts.

Fig. 3: Process of median filter A. Experiments with Noisy Background

In this part “monocot family plants” are included like rice,
In order to perform median filtering, first window is corn, wheat and iris with different background and without
moved and all the pixels enclosed by the window are shorted. background. In these plants disturbance is not introduced
After then median is computed and this value is assigned to because of vein. So disturbance is only introduced because of
center pixel. If the number of elements in K*K window is background and camera flash. Only brown spot diseases are
odd, middle value is assigned as median value, else average of included in this section.
two middle values is assigned as median value [5], [11]. Experimental results for disease spot detection of iris leaf
affected by heterosporium leaf spot disease using different
C. Disease Spot Segmentation
methods.
After image smoothing, a technique to detect the disease
spot is needed. It is important to select a threshold of gray
level for extract the disease spot from plant leaf. If the
histogram has sharp and deep valley between two peaks,
bottom of the valley can be chosen as threshold. But problem 1(a) 1(b) 1(c)
occurs when valley is flat and broad. In that case this
technique can’t be used to separate objects from background.
Therefore, Otsu method is used in this paper to automatically
select most desirable threshold [14]. 2(a) 2(b) 2(c)

OTSU Method
In OTSU method, the pixels are separated into two classes
C and " (background and object), using a threshold at level 3(a) 3(b) 3(c)
K. After then class means (DC , D" ) and class variances (EC , E" )
are calculated. Then a threshold K is searched, that maximizes
one of the object functions (l, k, n) [14].
GH% GL% GH%
F % ; K % ; 5 (13) 4(a) 4(b) 4(c)
GI GI GL%

III. EXPERIMENTATION
5(a) 5(b) 5(c)
In this research, images of rice, corn, wheat, iris, cotton,
soybean, mustard, magnolia, apple and cherry leaf are Fig. 4: RGB image of iris leaf affected by heterosporium leaf spot disease,
without background(1(a)), with noisy background(1(b)), with noise free
collected to find the best method for disease spot detection, background(1(c)) and their respective results of disease spot detection using
which is not affected by background and type of plant leaf. Method 1(2(a,b,c)), Method 2 (3(a,b,c)), Method 3 (4(a,b,c)) and Method 4
Four methods are discussed in this paper. (5(a,b,c)
 Piyush Chaudhary et al. 68

Experimental results for disease spot detection of rice leaf Experimental results for disease spot detection of Blueberry
affected by brown spot disease using different color transform leaf affected by bacterial canker (1(a)), frogeye leaf spot
methods. lesions on soybean leaf (1(b)), Bacterial blight leaf lesion on
cotton leaf (1(c)) using different color transform methods.

1(a) 1(b) 1(c)

1(a) 1(b) 1(c)

2(a) 2(b) 2(c)

2(a) 2(b) 2(c)

3(a) 3(b) 3(c)

3(a) 3(b) 3(c)

4(a) 4(b) 4(c)

4(a) 4(b) 4(c)

5(a) 5(b) 5(c)

5(a) 5(b) 5(c)
Fig.5 RGB image of brown spot disease on rice leaf, with noisy
background(1(a,b)), without background(1(c) and their respective results of Fig. 6 RGB image of Blueberry leaf affected by bacterial canker (1(a)),
disease spot detection using Method 1(2(a,b,c)), Method 2 (3(a,b,c)), Method frogeye leaf spot lesions on soybean leaf (1(b)), Bacterial blight leaf lesion on
3 (4(a,b,c)) and Method 4 (5(a,b,c) cotton leaf (1(c)) and their respective results of disease spot detection using
Method 1(2(a,b,c)), Method 2 (3(a,b,c)), Method 3 (4(a,b,c)) and Method 4
Through these results we can conclude that: (5(a,b,c)
• Using threshold on RGB image disease spot can’t be
detected accurately (method 1). Through these results we can conclude that:
• Using threshold on ‘H’ component of HSI color model • Using threshold on RGB image neither disease spot is
and ‘Cr’ component of YCBCR color model, disease not detected nor disturbance because of vein is
spots can be detected in some cases but not in all. So eliminated (method 1) (Fig. 6: 2(a, b, c)).
results are dependent on type of background (method 2, • Using threshold on ‘Cr’ component of YCbCr color
3). model, some disease spots are detected effectively, but
• Results show that using threshold on ‘A’ component of disturbance because of vein is present in results. So
CIELAB color model in all cases disease spots are results depend on type of leaf and vein (method2) (Fig.
detected accurately and results are independent of 6: 3(a, b, c)).
background (method 4) (Fig. 4,5 : 5(a,b,c)). • Using threshold on ‘H’ component of HSI color model
only few disease spots can be detected. Disturbance
B. Experiment with Disturbance of Vein because of vein is also present in some cases. So results
depend on type of leaf and background (method 3) (Fig.
In this part “dicot family plants” are included like cotton,
6: 4 (a, b, c)).
soybean, mustard, magnolia, apple and cherry. In these plants
• Using threshold on ‘A’ component of CIELAB color
noise is introduced because of longer and thicker veins.
model disease spots can be detected accurately in all
International Journal of Computer Science and Telecommunications [Volume 3, Issue 6, June 2012] 69

cases. Experiments show that results are independent of V. ACKNOWLEDGEMENT

type of plant leaf (method 4) (Fig. 6: 5(a, b, c)).
The authors sincerely like to thank the Agriculture
C. Experiment with Different Colored Disease Spots Research Station (A.R.S.), Fatehpur shekhawati (Rajasthan)
and VJTI. This work was not possible without help of Dr. S.
Till now it can be concluded that using CIELAB color
D. Singh, Dr. Vichitra Singh and all the people who have
transform (Method 4) noise is removed effectively, which is
directly and indirectly encouraged us and helped us in
generated because of background, camera flash and veins.
working out our research.
This method is also checked with different colored disease
spots and results are shown in Fig. 7.
REFERENCES

[1] Shen Weizheng, Wu Yachun, Chen zhanliang and Wei

Hongda3, “Grading Method of Leaf Spot Disease Based on
Image Processing”, International Conference on Computer
Science and Software Engineering, IEEE 2008, pp. 491-494.
[2] Santanu Phadikar and Jaya Sil, “Rice Disease Identification
using Pattern Recognition Techniques”, Proceedings of 11th
International Conference on Computer and Information
1(a) 1(b) 1(c) Technology (ICCIT 2008), Khulna, Bangladesh, IEEE, pp.
420-423.
[3] Nunik Noviana Kurniawati, Siti Norul Huda Sheikh Abdullah,
Salwani Abdullah, Saad Abdullah, “Investigation on Image
Processing Techniques for Diagnosing Paddy Diseases”,
International Conference of Soft Computing and Pattern
Recognition, 2009 IEEE, pp. 272-277.
[4] Song Kai, liu zhikun,Su hang,Guo chunhong ,“A Research of
2(a) 2(b) 2(c) Maize Disease Image Recognition of Corn Based on BP
Networks”, Third International Conference on Measuring
Fig. 7 : RGB image of frogeye leaf spot lesions on soybean leaf (1(a)),white
spot on mustard leaf (1(b)), grey spot on tomato leaf(1(c)) and their
Technology and Mechatronics Automation, 2011 IEEE, pp.
respective results of disease spot detection using Method 4 (2(a,b,c)) 246-249.
[5] Geng Ying, Li Miao, Yuan Yuan and Hu Zelin, “A Study on
the Method of Image Pre-Processing for Recognition of Crop
Result shows that different colored disease spots like black, Diseases”, International Conference on Advanced Computer
brown, gray and white colored disease spots can be detected Control, 2008 IEEE, pp. 202-206.
[6] Di Cui, Qin Zhang , Minzan Li, Glen L. Hartman and Youfu
accurately using method 4.
Zhao, “Image Processing Methods for Quantitatively
In some cases because of micronutrient deficiency and Detecting Soybean Rust from Multispectral Images”,
imbalance pigment formation, vein is differ in color in Published by Elsevier Ltd, Biosystems Engineering 107(2010),
comparison to plant leaf. In these cases disease spots can’t be pp. 186-193.
detected accurately using this method. [7] H. Al-Hiary, S. Bani-Ahmad, M. Reyalat, M. Braik and Z.
ALRahamneh, “Fast and Accurate Detection and
IV. CONCLUSION Classification of Plant Diseases”, International Journal of
Computer Applications (0975 – 8887), Volume 17– No.1,
In this paper YCbCr, HSI and CIELAB color models are March 2011
studied. All these color models are compared and finally ‘A’ [8] libo luo and Guomin Zohu, “Extraction of the Rice Leaf
Disease Image Based on BP Neural Network”, International
component of CIELAB color model is used. Color
Conference on Computational Intelligence and Software
transformed image is passed through median filter. In last, Engineering (CiSE 2009) ,IEEE
disease spots are segmented by applying OTSU threshold on [9] http://theseedsite.co.uk/monocots2.html, Accessed on april 28,
‘A’ component of LAB color space. 2012
Experimental result shows that noise which is introduced [10] http://www.backyardnature.net/mondiclf.htm, Accessed on
because of background, vein and camera flash; can be wiped april 28, 2012
out using CIELAB color model (Method 4). Following this [11] Rafeal C. Gonzalez and Richard E. Woods, “Digital Image
method different disease spots are detected accurately and Processing”, second edition, pearson education.
results are not affected by background, type of leaf, type of [12] compression.ru/download/articles/color_space/ch03.pdf
[13] AdrianFord and Alan Robert, “Color Space Conversions” ,
disease spot and camera.
available- http://www.poynton.com/PDFs/coloureq.pdf,
Further to this it is needed to compute disease spot area for Accessed on april 28, 2012
assessment of loss in agriculture crop. Disease can be [14] Nobuyuki Otsu, “A Tlreshold Selection Method from Gray-
classified by calculating dimensions of disease spot. In this Level Histograms”, IEEE transactions on systrems, man, and
work veins having color similar to the spot is not considered. cybernetics, VOL. SMC-9, no. 1, january 1979.
Further work need to be carried out in those lines also.
 Piyush Chaudhary et al. 70

Piyush Chaudhary was born in Alice Noble Cheeran was born in Kerala,
Rajasthan, INDIA on February 18, 1984. India on January 28, 1962. She received
He received his Bachelor degree in her B.E. in Electrical Engineering from the
Electronics and Communication Kerala University, India in 1984. She
Engineering from the Rajasthan joined the Electrical Engineering
university, in India in 2008 and he is department, Veermata Jijabai
currently pursuing M.Tech degree in the Technological Institute (VJTI), Mumbai,
department of Electrical Engineering, Veermata Jijabai India as lecturer in 1987. She completed
Technological Institute (VJTI), Mumbai, India. His research Masters in Electrical Engineering with specialization in control
interest is image processing (Email: systems in 1994 from Mumbai University, India. Subsequently
[email protected]). she completed Masters in Electronics Engineering in 1996
from the same university. Further did Ph. D. in the topic of
signal processing for hearing aids in 2005 from Indian Institute
of Technology, Bombay, India. Currently she is an Associate
Professor in the Electrical Engineering Department, Veermata
Anand Chaudhari was born in Jijabai Technological Institute (V J T I), Mumbai, India. Her
Maharashtra, INDIA on September 7, topics of interest include Signal and Image Processing
1987. He received his Bachelor degree Applications in various fields like biomedical, partial discharge
in Electronics and Telecommunication etc (Email: [email protected]).
Engineering from the Pune university, in
India in 2009 and he is currently pursuing
M.Tech degree in the department of
Electrical Engineering, Veermata Jijabai Technological Sharda Godara was born in Rajasthan,
Institute (VJTI), Mumbai, India. His research interest is image INDIA on August 5, 1990. She received
processing (Email: [email protected]). her Bachelor degree in Computer Science
from the MITS, Lakshmangarh in India in
2011 and she is currently working in
Accentur Services Pvt. Ltd., pune, India as
assistent software engineer. Her research
interest is image processing (Email:
[email protected]).