Measuring Leaf Area using Otsu Segmentation Method (LAMOS)

Size: px

Start display at page:

Download "Measuring Leaf Area using Otsu Segmentation Method (LAMOS)"

Jeffery Carter
5 years ago
Views:

1 Indian Journal of Science and Technology, Vol 9(48), DOI: /ijst/2016/v9i48/109307, December 2016 ISSN (Print) : ISSN (Online) : Measuring Leaf Area using Otsu Segmentation Method (LAMOS) Muhammad Haqqiman Radzali, Nor Ashikin Mohamad Kamal* and Norizan Mat Diah Department of Computer Science, Universiti Teknologi MARA, Shah Alam, Selangor, Malaysia; nor_ashikin@tmsk.uitm.edu.my Abstract Objective: This paper aims to measure the leaf area using image processing techniques that automate the grid counting method. Methods: For measurement of leaf area, firstly segmentation by Otsu method is required. Subsequently, denoising by median filter and followed by object recognition, boundary tracing and region filling techniques. The tool that is used in this study is Visual C Express using C++ and.net languages. Findings: Three types of leaves have been tested and the results show that this new method could be used in determining the leaf area with a small relative error. Application/Improvement: Leaf area is an important data to agronomist in conducting their research on plant growth, plant photosynthesis, plant physiological behavior and any other research that requires leaf area data. Keywords: Grid Counting, Image processing, Leaf Measurement, Segmentation 1. Introduction A leaf is one of the important parts of the plant that plays a very important role in plant photosynthesis and transpiration processes. One of the crucial parts that the agronomist always pays attention is the leaf area. By knowing the exact value of the leaf area, it will help researchers in examining the physiological features related to plant growth. The leaf area monitoring is one of the crucial aspects in examining physiological features concerning the plant growth, photosynthetic, and transpiration process. It is an important parameter in evaluating the damage caused by leaf diseases and pests, in order to discover micronutrients deficiencies, water and environmental stress, and the need for fertilization, for best management and treatment to the plant 1.According to one of the Betel (one type of herbs) producers, the features of Betel leaf such as the size and color are very important in categorizing the product in the market 1,2.In addition, the measurement of leaf physiological length and width are important features in the area of tea leaf recognition 3. Traditional methods such as grid counting, paper weighting, and leaf area meter are accurate but time consuming 4. In order to counter some of these problems, an image processing method has been selected to automate the grid counting technique into a new technique which is faster, more accurate, and easier to use to determine the size of a plant leaf. 2. Related Works Many methods have been introduced in measuring the leaf area such as planimeter, scanning, area-length regressions, grid counting, paper weighting, leaf area meter, and image processing methods. Planimeter is faster but limited in precision and with a high cost 1,5. The scanner is high accuracy but unable to constantly measure leaves of individual plants 6. Area-length regression is low in precision, uses an average principle and must get the coefficient of measurement by using another method first, then only the area can be calculated by measuring the length and width of the leaf 5. Grid counting, which is also known as square grid meter, uses a simple principle, takes more time, costs more materials, measures in vitro, can only measure leaves area, and high accuracy but sometimes may lead to low accuracy because it is easily affected *Author for correspondence

Measuring Leaf Area using Otsu Segmentation Method (LAMOS) by human subjective factors 1,6,7. Grid counting method takes more time and must be calculated in vitro 7.

In this technique, the leaf will be pressed on a paper, followed by outlining it with a pencil on the graph paper in which the smallest grid is measured as 1mm x 1mm.

Paper weighting is also known as gravimetric or copy weighting that uses a simple principle, takes more time, costs more materials, measures in vitro, can only measure leaves area, and high accuracy

In this technique, the leaf outline is cut out from the graph paper which is also called as a paper sample.

2 Measuring Leaf Area using Otsu Segmentation Method (LAMOS) by human subjective factors 1,6,7. Grid counting method takes more time and must be calculated in vitro 7. Besides the size of one grid in graph paper, grid counting fully depends on human observation and patience to calculate the area of the leaf. These existing methods are laborious. In this technique, the leaf will be pressed on a paper, followed by outlining it with a pencil on the graph paper in which the smallest grid is measured as 1mm x 1mm. After that, a total number of smallest grids within the outlined area which occupies more than 1/2 of the grid area will be calculated to get the area of the leaf 7. Paper weighting is also known as gravimetric or copy weighting that uses a simple principle, takes more time, costs more materials, measures in vitro, can only measure leaves area, and high accuracy but similar to the grid counting in which the measurement method is easily affected by human subjective factors 1,6. In this technique, the leaf outline is cut out from the graph paper which is also called as a paper sample. Then, this paper sample (that has been cut) weight (G1) is weighed in the electronic analytical balance. After that, the standard graph paper (10cm x 10cm) weight (G) is weighed in the electronic analytical balance and its area is identified (S=100cm 2 ). The paperweight per unit area is D=G/S, and the formula of leaf area is S1=G1/D 1,7. Leaf area meter is high accuracy, measures in vitro, more expensive, can measure leaf area, maximum length, and maximum width at the same time, repetition readings are essential, takes more time and energy when dealing with big leaves because it must be divided into segments in order to measure it and this will easily lead to errors 1,7. In this technique, leaf area, leaf maximum width, and length will be calculated using LI-3000A of LICOR Company according to leaves number. Each leaf is measured five times and the final result is the average of those five (5) measured values 1,7. The last method is an image processing method which is proven to be high accuracy, high precision, multiparameter, strong practicability, advanced technology, nondestructive measurement, faster, can measure in vitro not limited to the size of the leaf but can also measure the maximum length and maximum width 1,2,8,9. In this paper, LAMOS is implemented to measure the leaf area by using combinations of image processing algorithms. This project is developed and tested on a desktop computer with dual-core processor 2.3 GHz and 4 GB RAM. It uses Visual C Express as the platform to write the programming code which is in C++ and.net languages. This paper is organized into four sections. The methodology used is described in section 3. Experimentation and results are presented in section 4. The conclusionof the paper is presented in the last section. 3. Methodology Three (3) types of leaves have been selected to test the proposed method. These leaves are categorized into type A, type B, and type C with six (6), five (5), and (5) five leaves respectively. In this paper, leaf area is calculated using two methods that are grid counting method and LAMOS. Type A Type B Type C Figure 1. Image dataset. 3.1 Grid Counting Method Samples of leaves are traced on the 1 cm grid paper. Each occupied cell that is within the outlined area will be counted. A cell that occupies more than half outline will be counted as one (1) cell size. For example, let say one (1) cell is equal to 1cm 2, therefore the leaf area is equal to a number of occupied cells within the out lined area times the size of one cell 1,6,7. Figure 2 shows the grid counting method. The number of grid count corresponds to the actual area of the leaf. The leaf area in this method is calculated using Equation 1 below: Leaf area = NxB(1) Where N=Number of 1cm blocks covered by trace B=Area of one block in the graph paper Figure 2. Calculating the occupied cells within the outlined region 2 Indian Journal of Science and Technology

Muhammad Haqqiman Radzali, Nor Ashikin Mohamad Kamal and Norizan Mat Diah 3.2 Proposed Methodology

After the image has been loaded into the system, the first process is to convert the RGB image into a grayscale image as shown in Figure 5.

After the shrinking process, the system will then automatically convert the result of shrank image into a binary image by using Otsu segmentation method.

This process is important in order to correlate the leaf, reference object and background image.

3 Muhammad Haqqiman Radzali, Nor Ashikin Mohamad Kamal and Norizan Mat Diah 3.2 Proposed Methodology Steps of the proposed method (LAMOS) are shown below. Figure 3. Methodology. After the image has been loaded into the system, the first process is to convert the RGB image into a grayscale image as shown in Figure 5. If the image is bigger than the target size, it will be shrunkto a smaller size to minimize the time taken in conducting noise filtering. After the shrinking process, the system will then automatically convert the result of shrank image into a binary image by using Otsu segmentation method. The result of image segmentation is shown in Figure 6. Image segmentation is a process of grouping together pixels that have similar attributes. This process is important in order to correlate the leaf, reference object and background image. Some researchers use Histogram Color Threshold Approach, Adaptive Thresholding, Otsu segmentation and Gray Statistical Histogram for segmentation process and this paper opts for Otsu segmentation method. Since segmentation process depends on an image acquisition phase, there are times that Otsu segmentation does not segment the exact object successfully because of the error in the image acquisition phase. Thus, a slider is prepared to help the user to segment the object correctly. Image of the leaf is acquired using Sony Ericson K810i. A square of 1cm 1cm black drawing sheet is also captured with the leaf which will serve the purpose of a reference image in calculating the area. The captured leaf image will be saved as RGB image in JPEG format as shown in Figure 4. Figure 6. Binary image. Figure 4. RGB image. Sometimes, during the image acquisition, noise from the surrounding area may occur. LAMOS applies Median Filter technique to clean up the noise in the image taken. The higher the matrix size of the median filter, the higher the cleanliness of the image from the noise. Figure 7 shows the results of noise filtering by using median filter 7x7. Figure 5. Grayscale image. Figure 7. Median filter 7x7. Indian Journal of Science and Technology 3

Measuring Leaf Area using Otsu Segmentation Method (LAMOS) After removing the noise from the binary image, the next phase is to identify the leaf object and the reference object as shown in Figure 8.

color similar to the detected part of the object as in Figure 10.

Boundary tracing method is applied for edge detection of the leaf and reference object.

4 Measuring Leaf Area using Otsu Segmentation Method (LAMOS) After removing the noise from the binary image, the next phase is to identify the leaf object and the reference object as shown in Figure 8. Previously, on boundary tracing, it was done to help to complete a flood filling method (known as 4-neighbourhood style) in which the undetected part of the inner object will be filled up with a color similar to the detected part of the object as in Figure 10. After both objects have completed its region filling, now it is time to determine which one is the leaf object and which is the reference object depicted in Figure 11. Figure 8. Object recognition. Boundary tracing method is applied for edge detection of the leaf and reference object. In LAMOS, boundary tracing is required to fill up any inner region of the objects that supposedly to be part of the object. For example, if a leaf has some defects on its surface like holes caused by insect bites or different color from the healthy part of the leaf. In addition, the boundary tracing also adds the functionality of LAMOS in which the perimeter of the object could also be calculated. Other researcher had also used color features to detect leaves diseases 10. Figure 9 shows the result of boundary tracing. Figure 9. Boundary tracing. Figure 11. Leaf recognition. Below is the equation to calculate the leaf area using the proposed method: Leaf area = L2 x P (2) Where L = leaf area L2 = total number of pixel in leaf area R = reference object area R2 = total number of pixel in reference object area P = area for 1 pixel P = R2/R Equation 3 is used to calculate the relative error for each measurement in which the standard leaf area is an area that has been calculated by using grid counting method and measure leaf area is the area measured by using LAMOS method. Relative error = (measure leaf area - standard leaf area)(3) Standard leaf area 4. Results and Discussions Figure 10. Region filling (4 neighborhood styles). In analyzing the result, three types of leaves have been selected and categorized into type A, type B, and type C. Type A consists of six (6) leaves, five (5) leaves for type B and type C respectively. Table 1 shows the results of 4 Indian Journal of Science and Technology

5 Muhammad Haqqiman Radzali, Nor Ashikin Mohamad Kamal and Norizan Mat Diah Table 1. Result analysis Leaves Grid Counting Method Standard Leaf Area (cm 2 ) New Method (LAMOS) Measure Leaf Area (cm 2 ) Segmented Area(cm 2 ) Defect Area (cm 2 ) Reference Object Area (cm 2 ) Relative Error A A A A A A B B B B B C C C C C measuring leaf area by using Grid Counting Method and LAMOS. Both methods are being compared in terms of their total areas. LAMOS contains three columns namely; total area, segmented area, and defect area. The total area is a combination of the segmented area and defect area. The segmented area is the leaf part that is successfully segmented in Otsu segmentation process. This part may be considered as a healthy part of the leaf. The defect area, on the other hand, is the leaf part that is unsuccessfully segmented by Otsu segmentation process. It may be due to this particular part may be considered as an unhealthy part of the leaf or holes effect on the leaf. The reference object area used to test the results of both methods is 4 cm 2. It is found that the smallest relative error value when using LAMOS is whereas the largest relative error value is The large relative error value may be caused by the large size of the leaf area which makes it more difficult to count the occupied area using grid counting. Another reason may be due to the mistake done during image acquisition that makes the object and the background to appear similar in terms of the color and results in LAMOS to include some of the background areas as the leaf area. 5. Conclusion In this paper, we have implemented image processing algorithm for segmentation of leaf area. The proposed method was successfully applied to 3 types of leaf with a small relative error. The limitation of this application depends on the image captured during the image acquisition phase. The overall image must not be too bright or too dark and it must follow the criteria that are acceptable by the application, otherwise, it will make the application unable to segment the objects successfully and leads to error or failure in measuring the leaf area. In addition, the proposed method could also calculate the defect area of the leaf. 6. References 1. Patil SB, Bodhe SK. Betel leaf area measurement using image processing. International Journal on Computer Science and Engineering. 2011; 3(7): Soni AP, Dey AK, Sharma M. An image processing technique for estimation of betel leaf area. International Conference on Electrical, Electronics, Signals, Communication and Optimization (EESCO); p Indian Journal of Science and Technology 5

6 Measuring Leaf Area using Otsu Segmentation Method (LAMOS) 3. Jadon M, Agarwal R, Singh R. An easy method for leaf area estimation based on digital images. International Conference on Computational Techniques in Information and Communication Technologies; p Arunpriya C, Anthony ST. Fuzzy inference system algorithm of plant classification for tea leaf recognition. Indian Journal of Science and Technology Apr; 8(S7): Lu C, Ren H, Zhang Y, Shen Y. Leaf area measurement based on image processing. International Conference on Measuring Technology and Mechatronics Automation; p Feng T, Chun W. Calculating the leaf-area based on non-loss correction algorithm. Information Science and Management Engineering; p Tian YW, Wang XJ. Analysis of leaf parameters measurement of cucumber based on image processing. WRI World Congress on Software Engineering; p Gong A, Wu X, Qiu Z, He Y. A handheld device for leaf area measurement. Computer and Electronics in Agriculture. 2013; 98: Kaiyan L, JunHui W, Jie C, Huiping S. Measurement of plant leaf area based on computer vision. Sixth International Conference on Measuring Technology and Mechatronics Automation; p Padmavathi K, Thangadurai K. Implementation of RGB and grayscale images in plant leaves diseases detection-comparative study. Indian Journal of Science and Technology Feb; 9(6): Indian Journal of Science and Technology

Application of Machine Vision Technology in the Diagnosis of Maize Disease

Application of Machine Vision Technology in the Diagnosis of Maize Disease Liying Cao, Xiaohui San, Yueling Zhao, and Guifen Chen * College of Information and Technology Science, Jilin Agricultural University,