See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/326800397

Rice Leaf Blast Disease Detection Using Multi-Level Color Image Thresholding

Article · August 2018

CITATION READS

1 565

1 author:

mohd nazri abu bakar

Universiti Malaysia Perlis
3 PUBLICATIONS   9 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Rice Leaf Blast Disease Detection Using Multi-Level Color Image Thresholding View project

All content following this page was uploaded by mohd nazri abu bakar on 03 August 2018.

The user has requested enhancement of the downloaded file.

 Rice Leaf Blast Disease Detection Using Multi-
Level Colour Image Thresholding
M.N. Abu Bakar1, A.H. Abdullah2, N. Abdul Rahim2, H. Yazid2, S.N. Misman3 and M.J. Masnan4
1
Engineering Centre FTK, Universiti Malaysia Perlis
2
School of Mechatronics Engineering, Universiti Malaysia Perlis
3
Rice Research Centre, MARDI Seberang Perai
4
Institute of Engineering Mathematics, Universiti Malaysia Perlis
[email protected]

Abstract—Rice diseases have caused a major production and greyish green, later turning yellow to grey at the centre of
economic losses in the agricultural industry. To control and dead spots[3,4].
minimise the impacts of the attacks, the diseases need to be Several studies have reported that RLB disease is one of the
identified in the early stage. Early detection for estimation of most harmful diseases to rice and had caused yield losses. For
severity effect or incidence of diseases can save the production
from quantitative and qualitative losses, reduce the use of
instance, in Philippines and Japan, the yield losses were
pesticide, and increase country’s economic growth. This paper estimated to vary between 50% to 85% and 20% to 100%,
describes an integrated method for detection of diseases on respectively[5]. In 2014, [6] has reported that in Malaysia,
leaves called Rice Leaf Blast (RLB) using image processing nearly 10 thousand hectares of MADA crop area was attacked
technique. It includes the image pre-processing, image by this disease. This disease is capable of causing yield losses
segmentation and image analysis where Hue Saturation Value up to 100% which definitely affects farmers’ income and
(HSV) colour space is used. To extract the region of interest, national economy.
image segmentation (the most critical task in image processing) Traditionally, the RLB disease inspection is performed by
is applied, and pattern recognition based on Multi-Level trained human inspectors. However, their observations are
Thresholding approach is proposed. As a result, the severity of
RLB disease is classified into three categories such as infection
highly variable, and their decisions are always inconsistent
stage, spreading stage and worst stage. between inspectors from day to day basis [7]. Therefore, a
more reliable technique is necessary to diagnose this diseases
Index Terms—Rice Leaf Blast (RLB) Disease; Uncontrol instantly and accurately[8].
Environment; Image Pre-processing; Colour Image Thus, an early detection system has been proposed in the
Segmentation; Multi-level Image Thresholding. literature based on image processing technique[8-10].
Although human eye can identify some of these symptoms,
I. INTRODUCTION image processing system gives more accuracy and
speed[11,12]. This technique will be used to reduce manual
Rice plants can become stressed mainly due to the abiotic and inspection and identification of common rice disease[13,14].
biotic factors. Abiotic is the non-living factors such as wind Unfortunately, the existing image processing system is
and water, while other living organisms like plants, animals, focused only on a single leaf and unfavourable for many other
bacteria and fungi are called the biotic[1]. Rice plants’ aspects such as different sizes, orientations, complex
stresses are expressed in various ways. For example, plant background and light conditions. It is very challenging to
water stress can slow down photosynthesis, reduce detect the target image outdoor since it composed both the
evapotranspiration, and raise leaf surface temperature. Other structured and unstructured objects [15].
symptoms can be caused by bacteria, viruses, or fungus which This work presents RLB disease diagnosis system using
include morphological changes like discolouration of leaves, multi-level colour image thresholding. The system includes
wilting and plants that do not produce panicles or produce image enhancement, image segmentation, and feature
empty panicles. extraction. Experimental work was set up at a rice field with
Rice Leaf Blast (RLB) disease is one of the biotic stresses considering many rice leaves, different sizes, orientations and
that contribute to the reduction of rice productivity light conditions. Details of the proposed system are shown by
worldwide. It has become a dominant, threatening disease in the flowchart in Figure 1.
Malaysia due to its wide distribution and ability to survive in
a wide range of environmental conditions. II. METHODOLOGY
The RLB disease is mainly caused by fungus Pyricularia
oryzae Cavara[2] that attack rice plants which diminish rice The experiment was conducted at Rice Research Centre of
quality. This disease can attack all stages of rices’ growth as Malaysian Agricultural Research and Development Institute
early as the seedlings, vegetative and even the harvesting (MARDI) Seberang Perai, Malaysia. The acquired colour
stages. The experts can identify the symptoms of this disease, images were captured by using smartphone camera in real-
i.e. by looking at a leaf with lesions usually starts near the leaf world conditions where the condition of crops’ leaves and the
tips or leaf boarder or both. These symptoms appear as brown environmental conditions are uncontrolled. Then, all the
specks and then grow to become spindle-shaped, pointed at images will be transferred to a computer for detection of RLB
both ends. The colour of lesions is usually pale green to by using image processing technique. The output expected
data such as the severity stages of RLB disease can be used

e-ISSN: 2289-8131 Vol. 10 No. 1-15 1

Journal of Telecommunication, Electronic and Computer Engineering

for further action by farmer or agricultural experts. This RLB

disease detection system is shown in Figure 2. 𝑃 = 𝑃$ + 𝑛 (1)

where 𝑃$ is the true value of pixel and 𝑛 is the number of

Start noise in that pixel. These noisy pixels need to be removed to
produce a visually high quality image before further process
of retrieving the original images. Ideally, after image
Image Acquisition restoration, the new image should get;

𝑃 = 𝑃$ (2)
Image Pre-processing
*Resize where the mean of noise is zero. Image restoration or
*Enhancement denoising includes the process of changing, correcting, or
*Noise Restoration moving of the image data to produce noise-free image while
preserving the edges. This step should be processed before
Image Segmentation
the segmentation stage without loss of information for RLB
*Background Subtraction disease diagnoses to avoid from false feature selection during
*Feature Extraction background subtraction.
*Image Analysis
C. Image Enhancement
This process will improve the low intensity of original
Experimental Result image by reducing its dynamic range and/or increasing its
contrast. This process will enhance the quality of the image
that suits the image processing analysis. The Histogram
End
Equalization (HE) method is used to enhance the contrast of
the acquired image. This method will be equalising the
Figure 1: The system flowchart
intensity distribution of an image or flattening the intensity
distribution curve. This method is the most useful method for
enhancing the quality of image[16-18]. The original Red,
Green and Blue (RGB) colour image is transformed into Hue
Saturation Value (HSV) colour space.

(a) (b)
Figure 2: The developed rice blast diseases detection system.

III. IMAGE PRE-PROCESSING

The colour images need to enhance before it can be

analysed by the pattern recognition technique. The image pre-
processing is used where the original images will be
transformed into new colour space, This new image (c) (d)
fundamentally similar to the source image, but differs in
certain aspects. This process involves image resize, image Figure 3: (a) Original RGB image. (b) V channel before enhancement.
restoration (filtering) and image enhancement. (c) After enhancement. (d) Convert back to RGB.

A. Resize Then, it will be split into a single component which Hue

The original images will be resized to a fixed resolution to (H), Saturation (S) and Value (V) from HSV respectively.
improve the memory storage capacity and to reduce the Only V component is enhanced while other components
computational complexity. The resizing image will reduce to remain in its coordinates. After enhancement, the new V
the dimension of 640 x 480 pixel. component is merged back to it colour space. Figure 3(a) and
Figure 3(b) shows the original image and enhancement
B. Noise Restoration process for V channel. Figure 3(c) and Figure 3(d) shows the
Noise is one of the undesirable nuisances in the acquired result after merging and convert back to RGB colour space.
images. It may be caused by motion between camera and
object, improper shutter opening, atmospheric disturbances IV. IMAGE SEGMENTATION
and misfocusing [19]. Generally, it can be considered as a
random variable with zero mean. Consider a noisy pixel; Image segmentation is a process to divide an image into
significant regions or segments and extract out the interested

2 e-ISSN: 2289-8131 Vol. 10 No. 1-15

 Rice Leaf Blast Disease Detection Using Multi-Level Colour Image Thresholding

target used for image processing. The process will partition a B. Feature Extraction
colour image into discrete regions which contain information The image analysis that based on edge detection is
which generally very challenging. The suitable segmentation unaffected to change in the overall illumination level[23].
technique depends on the type of image and problem facing. From Figure 5, the image result is observed as in the
Every region has similar pixel according to some disconnected white regions which influence the decision in
homogeneity image characteristics such as colour, intensity image interpretation. Therefore, the regions of RLB disease
or texture[20]. in the image should be completely solid before the image
Image segmentation is an essential step for image suitable to analyse. A morphological closing is performed by
interpretation, analysis process, object representation and using a 7 x 7 rectangular structuring element to join all the
visualisation. The colour image segmentation technique s disconnected regions together as shown in Figure 5(a).
include pixel-based, edge-based, region-based, model-based,
physics-based, and section surveying hybrid-based[22].
There are no rules in selecting the image segmentation
technique suitable for an image. It is also difficult to select a
various approach for a specific type of image. Most
researchers will diligently try to resolve the uncertainty
difficulties encountered while trying to model the human
visual system[21]. Therefore, selecting the suitable image
segmentation technique remains a challenging problem in (a) (b)
image processing[22].
The image segmentation process consists of, background
subtraction, feature extraction and image analysis.

A. Background Subtraction
The objective for background subtraction is the region of
leaf affected by RLB disease and define as a foreground.
Segmentation is performed by choosing suitable thresholds
range to extract the infected area from the background.
(c) (d)
Thresholding is an important method for image
segmentation which is based on image space regions, i.e. the
characteristics of the image. This image segmentation
technique has an advantage compared to others due to it is
smaller storage space, fast processing speed and easy
manipulation[20].
The values of the threshold are chosen from the lower and
upper of the image histogram. Since the colour of RLB
diseases region is not uniformly distributed, a single process
of thresholding is not efficient. Therefore, multi-level (e)
thresholding based on pixel-based segmentation techniques Figure 4: (a) Original input Image. (b) 1st level thresholding. (c)2nd level.
for a colour image is proposed. This is illustrated by Equation (d) 3rd level. (e) Result of total image thresholding.
(3)

𝑣0 , 𝑖𝑓 𝑇0 ≤ 𝑇 (𝑥, 𝑦) < 𝑇5
𝑇( (𝑥, 𝑦) = .𝑣5 , 𝑖𝑓 𝑇 (𝑥, 𝑦) ≥ 𝑇5 (3)
𝑣7 , 𝑖𝑓 𝑇(𝑥, 𝑦) ≤ 𝑇0

where 𝑇0 and 𝑇5 is the lower and upper range of thresholding

values, respectively. A point (x,y) is classified as object of
class 1 (𝑣0 ) if the threshold value is between 𝑇0 and 𝑇5 . If the
(a) (b)
threshold value is greater than or equal to the upper range 𝑇5 ,
a point (x,y) is classified as object of class 2 (𝑣5 ), and it is Figure 5: (a) Example of disconnected regions image. (b) Canny edge
classified to the background if the threshold is not more than detection after applied Gaussian Smoothing filter.
the lower range 𝑇0 . The HSV is used according to the rate of
high luminance points and white luminance points in the Edge detection is used in most image analysis which the
image. outlines of the object in the image will detect the boundaries
The H channel is selected for its ability to accommodate the between regions of objects and background. The canny edge
variable lighting conditions and to distinguish between rice detector is selected since it produces thin lines, continuous
and non-rice leaf colour efficiently. For every pixel, the H edges and much better detection compare to others algorithm
channel will be checked that allow only values above a especially in various lighting conditions. Figure 5(b) shows
specific range to be realised based on the selected threshold an example of edge detection after applying 5 x 5 Gaussian
value. Figure 4 shows the example of RLB disease colour smoothing filters that will reduce speckle of high-frequency
extraction using this technique. noise.

e-ISSN: 2289-8131 Vol. 10 No. 1-15 3

Journal of Telecommunication, Electronic and Computer Engineering

C. Image Analysis
Table 1
Then the image contour and region boundary of interest
The Experimental Result of Rice Disease Detection.
will be trace. A contour is defined as a segment that is one
pixel wide, and one or more pixels in length and a boundary No. of detected Stage of
Area in Pixel
is an unbroken contour[24]. This feature extraction finding disease region Severity
represents the region of RLB disease. The border tracing 1 172.5 2
algorithm is used to extract the contours. Figure 6(a) shows 2 27 1
the result of this disease detection where the area of infected 3 333.5 2
are marked with red colour. Then, the infected area of the 4 50.5 1
5 705.5 2
disease is calculated using the Green’s theorem as in the
6 99 1
following [25];
7 14.5 1
8 21.5 1
=> =A 9 81 1
∫F 𝑃𝑑𝑥 + 𝑄𝑑𝑦 = ∬E < =? − =B C 𝑑𝐴 (4)
10 1762.5 3
11 16 1
12 2398.5 3
where 𝐶 is positively oriented a simple closed curve in the
13 16 1
plane, 𝐷 the region bounded by 𝐶, and 𝑃 and 𝑄 are functions 14 1108 3
with continuous partial derivatives in an open region 15 95 1
containing 𝐷. 16 443.5 2
The RLB disease infected area in an image is based on the 17 384.5 2
calculated boundary region as shown in Table 1. It also shows
the suggested three-stage disease severity. Figure 6(b) Table 2
illustrate the suggested severity stage; stage one is for early The Experimental Results in a Varoius Uncontrolled Enviroment.
infection phase (red), stage two is for spreading phase (blue), No. of
and stage three is for worst phase (green). Image Stage Stage Stage
detected
No. 1 2 3
disease region
1 0 0 0 0
2 36 14 22 0
3 8 5 3 0
4 13 6 7 0
5 28 16 9 3
6 3 2 1 0
7 36 19 16 1

VI. CONCLUSION
(a) (b)
This paper has discussed a technique to detect RLB disease
Figure 6: (a) Boundary of rice disease region on leave. (b) The severity using multi-level colour image thresholding. The technique
stage. successfully detects the disease based on the images taken in
uncontrol environment. The technique is not suitable for
V. FIELD EXPERIMENT detection of other diseases which may have similar features.
However, this technique can be further enhanced by having
Figure 7 shows seven more experiments of RLB disease the capabilities to measure the incidence of RLB disease in
detection where the first row is the input image captured the field.
under uncontrolled environment, i.e. multiple rice leaves, the
different sizes, orientations, light conditions and complex ACKNOWLEDGEMENT
background.
The first row shows various background conditions in This work is conducted with collaboration by Rice
uncontrol environment of the captured images due to Research Centre of Malaysian Agricultural Research and
uncontrolled environment. The first column from left shows Development Institute (MARDI) Seberang Perai. M.N. Abu
the healthy rice leaf and the other column of the images Bakar also gratefully acknowledges Universiti Malaysia
consisting unhealthy rice leaf. The second and third row Perlis (UniMAP) for the opportunities given to conduct the
shows the result of image transformation and pre-processing research.
stage, respectively. The fourth and fifth row illustrates the
image segmentation and analysis stage. The sixth row shows
the result of image analysis where the region of RLB disease
boundaries with red in colour. The final row illustrates the
result of this disease detection which the severity stage of
each image is summarised in Table 2.

4 e-ISSN: 2289-8131 Vol. 10 No. 1-15

 Rice Leaf Blast Disease Detection Using Multi-Level Colour Image Thresholding

1) 2) 3) 4) 5) 6) 7)

Figure 7: The results of seven experiments (by column) shows a different condition and each row represents intermediate results of image processing step
for the proposed method. From left to right: healthy rice plant, image include another object, the image captured from the top, focus image, dark image, blur
image and high intensity image. (These figures are best viewed in colour)

REFERENCES [11] R. A. D. Pugoy and V. Y. Marianoa, "Automated rice leaf disease

detection using color image analysis," in Third International
Conference on Digital Image Processing, 2011.
[1] S. Phadikar, A. Das and J. Sil, "Classification of Rice Leaf Diseases
[12] C. -L. Chung, K.-J. Huang, S.-Y. Chen, M.-H. Lai, Y.-C. Chen and Y.-
Based on Morphological Changes," International Journal of
F. Kuo, "Detecting Bakanae disease in rice seedlings by machine
Information and Electronics Engineering, vol. 2, 2012.
vision," Computers and Electronics in Agriculture 121 (2016) 404–
[2] C.M. Vera Cruz, I Ona, N.P. Castilla and R. Opulencia,
411, vol. 121, pp. 404-411, 2016
http://www.knowledgebank.irri.org, 2016.
[13] G. Bhadane, S. Sharma and V. B. Nerka, "Early Pest Identification in
[3] https://blogmardi.wordpress.com/2017/02/07/blb-kurangkan-berat-biji
Agricultural Crops using Image Processing Techniques," International
rin-padi, 2017
Journal of Electrical, Electronics and Computer Engineering , vol. 2,
[4] G. Miah, M. Y. Rafii, M. R. Ismail, A. B. Puteh, H. A. Rahim,
no. 2, pp. 77-82, 2013.
R. Asfaliza, M. A. Latif, “Blast resistance in rice: a review of
[14] A. Vibhute and S. K. Bodhe, "Applications of Image Processing in
conventional breeding to molecular approaches”, Molecular Biology
Agriculture: A Survey," International Journal of Computer
Reports, Vol. 40, Issue 3, pp 2369–2388, 2013.
Applications , vol. 52, no. 2, pp. 0975-8887, 2012.
[5] C.J. Chuwa, R.B. Mabagala2, M. S.O.W. Reuben,” Assessment of
[15] E. S. Sama and A.K. Prabhavathy and J.D. Shree, " A Survey on
Grain Yield Losses Caused by Rice Blast Disease in Major Rice
Outdoor Scene Image Segmentation," International Journal of
Growing Areas in Tanzania”, International Journal of Science and
Computer Applications, vol. 55, no. 9, 2012.
Research, Vol. 4, Issue 10, 2015.
[16] J. Purushothaman, M. Kamiyama and A. Taguchi, "Color image
[6] http://madainfo.blogspot.my/2015/02/taklimatkepadamediaberkaitanp
enhancement based on Hue differential histogram
erosak.2017.
equalization," International Symposium on Intelligent Signal
[7] S. Sankarana, A. Mishraa and R. Ehsania, "A review of advanced
Processing and Communication Systems (ISPACS), Phuket, pp. 1-5,
techniques for detecting plant diseases," Computers and Electronics in
2016.
Agriculture, vol. 72, pp. 1-13, 2010.
[17] T. K. Agarwal, M. Tiwari and S. S. Lamba, "Modified Histogram based
[8] S. Phadikar , J. Sil and A. K. Das, "Rice diseases classification using
contrast enhancement using Homomorphic Filtering for medical
feature selection and rule generation techniques," Computers and
images," 2014 IEEE International Advance Computing Conference
Electronics in Agriculture, vol. 90, p. 76–85, 2013.
(IACC), Gurgaon, 2014, pp. 964-968.
[9] J. G. A. Barbedo, "Digital image processing techniques for detecting,
[18] R.Garg, B.Mittal, S.Garg,” Histogram Equalization Techniques for
quantifying and classifying plant diseases," Springerplus, vol. 2, no. 1,
Image Enhancement”, International Journal of Electronics &
pp. 660-671, 2013.
Communication Technology, Vol. 2, Issue 1, 2011.
[10] E. Singh and M. L. Singh, "Automated Colour Prediction of Paddy
[19] Archana, Sheenam, A. Cchabra, “Comprehensive Review of Denoising
Crop Leaf using Image Processing," in IEEE International Conference
Techniques in Image Restoration”, International Journal of Innovative
on Technological Innovations in ICT for Agriculture and Rural
Research in Computer and Communication Engineering, Vol. 2, Issue
Development, 2015.
6, June 2014.
[20] W.Khan, “Image Segmentation Techniques: A Survey”, Journal of
Image and Graphics, Vol. 1, No. 4, 2013.

e-ISSN: 2289-8131 Vol. 10 No. 1-15 5

Journal of Telecommunication, Electronic and Computer Engineering

[21] R.Harrabi and E.B.Braiek, “Color image segmentation using multi- [23] M.S. Nixon, A.S. Aguado, “Feature Extraction and Image Processing”,
level thresholding approach and data fusion techniques: application in Second edition 2008.
the breast cancer cells images”, Journal on Image and Video [24] W.A. Mcqueen, “Contour tracing and boundary detection for object
Processing, 2012. identification in a digital image”, U.S. Patent 6674904, 2004.
[22] R.Dass, Priyanka, S.Devi, “Image Segmentation Techniques”, [25] https://www.whitman.edu/mathematics/calculus_online/section16.04.
International Journal of Electronics & Communication Technology html.
,Vol. 3, Issue 1, 2012.

6 e-ISSN: 2289-8131 Vol. 10 No. 1-15

View publication stats