CANNED PINEAPPLE GRADING USING PIXEL COLOUR EXTRACTION

Transcription

1 CANNED PINEAPPLE GRADING USING PIXEL COLOUR EXTRACTION Sharmiza Kamaruddin 1, Sunardi 2, Kamarul Hawari Ghazali 3, Rosyati Hamid 4 Department of Electrical Engineering, Politeknik Sultan Haji Ahmad Shah, Pahang, MALAYSIA 1 sharmiza@polisas.edu.my Faculty of Electric and Electronic Engineering, University Malaysia Pahang, MALAYSIA 2 sunardi@ump.edu.my 3 kamarul@ump.edu.my 4 rosyati@ump.edu.my Abstract Malaysia is a one of the main producer of canned pineapple in the world. According to Malaysian Pineapple Board Industry, it is about 5% of total production of canned pineapple is meant for export. MPIB was responsible in controlling the quality of product before exporting abroad. Quality inspection of the pineapple was done manually by quality inspector. The grade is depending on the color of the pineapple and producers need to get certification of grade before they can export to overseas. In this paper, we presented our study on automatic detection of canned pineapple grade using image processing technique. MPIB currently has done the quality inspection using expertise worker which is currently not suitable to be applied. This research develops to differentiate and classify the standard 15 and standard 16 of canned pineapple. The image data collections were controlled in order to avoid any outside lighting source. Then, Otsu s method, morphological operation has been applied on the image to make sure the region of interest (ROI) quality is the best before multiplying it with original image. Using Red Green Blue (RGB), Hue Saturation Value (HSV) and CieLAB colour space the pixel colour were extracted using mean and standard deviation. 100% classification between two standard of canned pineapple was determined using Hue component in HSV colour space. Keywords: Image Processing, Colour Space, Canned Pineapple 1. Introduction In the 21 st century there are a lot of systems that being automatically from manual. This is due to demand on fast, efficient, quality and cost either from manufacturer or individual person. It was applied on agriculture, transportation, financial or other industrial as a part of solution for their problem. In agriculture industry, some of automatically system was using image processing technique such as fresh pineapple [1, 2], colour scale for pear [3], star fruits [4], papaya [5], tomato [6], palm oil [7], and apple [8]. This research was focused on canned pineapple to be use with image processing techniques. The pineapple canning industry was the main contributor of the Malaysia s economy and the second highest export after watermelon in the tropical fruit category. According to Malaysian Pineapple Industry Board (MPIB) 95% of canned pineapple productions are for export market and 5% is for domestic market while fresh pineapple contributes 3% to export market and 70% to domestic market. This industry played a role in contributing to the country s economy and provides direct job opportunities in the plantation and processing industry and indirectly in the transportation and manufacturing industry. Malaysia is capable of producing high quality product that can survive in the mainstream market [9]. Malaysia Pineapple Board Industry (MPIB) was responsible in controlling the quality of industry manufactured products and provide guarantee of safety either in processed form, semi-process, or fresh form. Organized by WorldConferences.net 217

2 There are a lot of quality inspection done in MPIB lab such as food chemical analysis service, proximate analysis service, microbiology analysis service, and grading analysis service. In grading analysis service, there are focusing on canned pineapple quality inspection. Figure 1 show the canned pineapple process before being export. One of the quality inspections is colour of canned pineapple before being export to several countries. However this quality colour inspection is done by experience worker using their eye recognition. This manual method will cause a different result for the quality inspection since they will ignore the environment condition such as weather and light. Manual inspection also cause to labour intensive, slow and can be inconsistent due to fatigue and due to the relatively large staff turnover caused by boredom [10]. A research have been done using different colour space such as on star fruit colour maturity classification which is based on RGB colour space convert into YCbCr colour space[11], differentiate oil palm bunch based on RGB intensity [12],segmentation on colour image using HSV colour space[13] and also starfruit classification based on linear hue[14]. Image processing or colour identification is one of the method to grad the suitable canned pineapple either cube or slice. Therefore, image processing that using colour to differentiate and grad according it standards will be used in this research as a main contribution. We will use different colour spaces were used such as RGB, HSV and Cielab to get most suitable canned pineapple grading. Fruit Receiving Fruit Cleaning Manual Fruit Inspection & classification Fruit peeling & slicing Filling & Syruping Storage Cooling & Dryng Sterilizing Seaming Exhausting Quality Inspection- Colour, Sugar Export Figure 1: Canned Pineapple Process 2. Methodology Five steps were taken to achieve the goal of this research. It was start from data collection, image preprocessing, features colour extraction, classification and followed by decision. Using MATLAB R2010b, each step was followed accordingly with image processing toolbox since it has a collection of functions that extend the capability of the MATLAB numeric computing environment. Figure 2 show the Diagram of research methodology. 2.1 Data Collection /Acquisition In image processing, data acquisition is important to get the image that can help minimize a certain error that might occurred during first processing of image processing until the final decision. All images were taken under supervision of expertise worker in MPIB to make sure the canned pineapple data in the correct standard. Organized by WorldConferences.net 218

3 Data Collection/Acquasition Image preprocessing Features Extraction Decision Classification Figure 2: Diagram of research methodology The data of canned pineapple being export were fall in standard 15, standard 16 standard 17 and standard 18. However, in this research we were focus on standard 15 and standard 16 because, normally this two s standard will be sent from factory to MPIB laboratory. a b Colour of a lighter than b without environmental control Figure 3: Image Acquisition first phase In the first phase, the images were taken using Canon G12 digital camera, with same resolution 12 Megapixel and same height of tripod. During the snapshot, only room light has been used instead of light source from nearest window. The illustration of image acquisition in the first phase was as Figure 3. Figure 4 shows the image of canned pineapple in the first phase. Figure 4: Image of canned Pineapple (first phase) Second phase of image acquisition, the image was taken under control environment using cool white light with luminos flux 600Lm. Logitech webcam with version 2.31 software were using while data snapshot and the image resolution 640x480 pixels to decrease the processing time during simulation. The webcam was place on the top of box which is no other light source will come in except from the light as mention above. Different light exposure will generate different result [14]. Canned pineapple was place on the blue background to make sure we get the clear image since in RGB colour space yellow colour in blue component was less compared to other component. Image histogram in Figure 3 Organized by WorldConferences.net 219

4 proves the yellow component in RGB colour space. Figure 4 show the image acquisition in second phase. While, Figure 5 show the examples of canned pineapple image in the second phase. Figure 4: Image acquisition in second phase. 2.2 Image Pre-processing Figure 5: Canned pineapple image in the second phase. Image pre-processing is the method to get the image of canned pineapple and eliminate others unwanted pixel value from image such as blue background image. Figure 6 show the block diagram of pre-processing process in this research. Image 1 Otsu's Methodthreshold Morphological Process Class converting Multiplying mask with Original Image Figure 6: Block diagram of pre-processing process Otsu s Method is a global threshold level computes used to convert an RGB or intensity image to a binary image. Since Binary image was refer to pixel value 1 and 0, level will normalize the intensity value that lies in the range [0, 1]. The graythresh function in MATLAB uses Otsu's method, which chooses the threshold to minimize the intraclass variance of the black and white pixels [15]. Figure 7 and 8 show the different binary image using Otsu s method and without Otsu s method. In order to eliminate unwanted pixel, Otsu s method was chosen before applying morphological process. Figure 7: Binary Image with Otsu s Method Organized by WorldConferences.net 220

5 Figure 8: Binary Image without Otsu s Method Morphological image processing is a collection of non-linear operations related to the shape in an image. Morphological operations rely only on the relative ordering of pixel values, not on their numerical values, and therefore are especially suited to the processing of binary images. It was a process to eliminate unwanted pixel in the image and to focus on region of interest (ROI). Dilation, erosion, opening and closing is the operation that always being use in image processing. Dilation is the basic effect of the operator on a binary image to gradually enlarge the boundaries of regions of foreground pixels. Thus areas of foreground pixels grow in size while holes within those regions become small. While, erosion is to erode away the boundaries of regions of foreground pixels. Thus areas of foreground pixels shrink in size, and holes within those areas become larger. Mathematically, with A and B as sets in Z 2, the dilation of A (image) by B is defined as ( ) (1) And With A and B as sets in Z 2, the erosion of A by B is defined as ( ) (2) Dilation and Erossion was applied to get the region of interest (ROI) which is in this research refer to the canned pineapple. Figure 9 and Figure 10 was the example of ROI before and after goes through dilation and erosion process. The final binary image after morphology process was called masking image and it will multiply with original image of canned pineapple. The image of canned pineapple was showed in Figure 11. Figure 9: Image before morphology operation Figure 10: Image after morphology proses Organized by WorldConferences.net 221

6 Figure 11: Image of canned pineapple 2.3 Features Extraction Yellowish canned pineapple was different between each standard of canned pineapple. A yellowish colour scale has been developing as [16], however the pixel colour in standard 15 also can exist in standard 16. It was recommended that the two colour scale to be use automatically with computer vision rather than print it into a paper because we must consider the printer s features colour. To overcome this problem, we work on mathematical statistic with different colour space such as RGB, HSV and CieLAB RGB colour space In RGB colour space, an imported image on a computer is thus transformed into 3 matrices with values per pixel to represent the main colour. Pure Yellow in RGB colour space was [255,255,0]. Each pixel in canned pineapple represent different value and blue component in RGB contribute the smallest value for yellowish colour. Figure 12 show the pixel region of canned pineapple, while Figure 13 the image histogram of red, green and blue component in canned pineapple with y axis representing the colour concentration and x axis pixel value. Figure 12: Pixel Region of canned Pineapple Figure 13: RGB Image Histogram By referring to a research in star fruit [14], blue component was ignoring because of the less contribution in star fruit yellowish colour. Meanwhile, based on Figure 12 although blue component contribution is less, it differentiates the yellow colour of each pixel. Statistical values on mean and Standard Deviation (STD) were calculated for blue component in RGB colour space involving both standard of canned pineapple. Table 1 show the result of RGB colour space for Mean and STD. Organized by WorldConferences.net 222

7 Table 1: Mean and STD RGB colour space Statistic RED GREEN BLUE S15 S16 S15 S16 S15 S16 Mean Standard Deviaton (STD) HSV Colour Space In Hue Saturation value (HSV), H component represents color information, can provide more obvious characteristics of and can also be used as the semantic features of images to the image classification. Saturation is the intensity of the colour and ranges from 0 to 100% while value is the brightness of the colour. The advantages of HSV over RGB are: a) Hue is invariant to certain types of highlights, shading, and shadows; b) The statistic value is performed on only one dimension (H) and results of statistic have fewer than using RGB [16]. Table 2 show the result of HSV colour space for Mean and STD. Mean and STD in HSV colour space cannot distinguish between two standard of canned pineapple especially for Value or also known as brightness. The conversion formula that transfers from RGB to HSV is defined as below: ( ) ( ) ( ) ( )( ) (3) ( ( )) (4) ( ) (5) Table 2: Mean and STD HSV colour space. Statistic HUE Saturation Value S15 S16 S15 S16 S15 S16 Mean Standard Deviaton (STD) CieLAB Colour Space In CieLAB colour space, one channel is for Luminance (lightness) (L) and another two colour channels (a and b). The colour differences which we perceive correspond to distances when measured colorimetrically. The a axis extends from green (-a) to red (+a) and the b axis from blue (-b) to yellow (+b). The brightness (L) increases from the bottom to the top of the three-dimensional model. The maximum for L is 100, which represents a perfect reflecting diffuser and minimum is 0, which represent as black. Table 3 showed the value of Mean and STD using CieLAB colour space.mean and STD in a* and b* between standard 15 and standard 16 also too close meaning that we cannot simply take the value on Table 3 to make a decision in this research. Table 3: Mean and STD using CieLAB colour space Statistic L a* b* S15 S16 S15 S16 S15 S16 Mean Standard Deviaton (STD) Organized by WorldConferences.net 223

8 3. Result and Discussion Table 4 and Table 5 show the percentage of classification using different structuring element (SE) according to Mean and STD on each colour space. There were not too much different between each SE because of the parameter we use were related with a little changes depend on type of shape. Table 4: Mean Percentage of Classification with different SE SE RGB LAB HSV R G B L A B H S V DISK DIAMOND LINE Table 5: STD Percentage of Classification with different SE SE (shape) RGB LAB HSV R G B L A B H S V DISK DIAMOND LINE SQUARE ARBITRARY SQUARE ARBITRARY Refer to our research and theoretical of RGB, HSV and CieLAB colour space, blue component in RGB contribute the percentage in yellowish colour, Hue represent the yellow colour without intensity and brightness, and b+* is for yellow colour. Statistics for the three components have been test and Figure 14, Figure 15 and Figure 16 was the graphs mean versus number of image. Figure14: Mean Pixel Value b (RGB) Plane Vs No of Sample Organized by WorldConferences.net 224

9 Figure15: Mean Pixel Value b* Plane Vs No of Sample Figure16: Mean Pixel Value Hue Vs No of Sample Blue component in RGB image and Hue give 100% of classification however b* (CieLAB) only give us 93% of classification and it same goes to STD. 4. Decision and Conclusion Every value of pixel in an image means a lot when we did a research using image processing especially in colour techniques. In RGB colour space, a yellowish colour was depend on each component contribution, while Hue was an attribute of a visual sensation according to which an area appears to be similar to one of the perceived colours as in this case yellowish canned pineapple. Although both blue components in RGB and hue give 100% classification, HSV colour space were choose as a method of classification in this research. In addition, HSV are simple transformations of device-dependent RGB models, the physical colours define depend on the colours of the red, green, and blue primaries of the device or of the particular RGB space, and on the gamma correction used to represent the amounts of those primaries. Organized by WorldConferences.net 225

10 References Rohana Abdul Karim, Kamarul Hawari Ghazali, Nurul Wahidah Arshad, Nor Farizan Zakaria, Nazriyah Bt Che Zan, Pineapple Maturity Inspection using Colour Identification, International Conference on Instrumentation, Control & Automation, ICA2009. Shuhairie Mohamad, Kamarul Hawari Ghazali, Nazriyah Che Zan, Siti Sofiah Mohamada Radzi, Rohana Abdul Karim, Classification of Fresh N36 Pineapple Crop using Image Processing Technique, ICENC Yoshitaka Motonaga, Tatsuya Matsumoto, Naohiko Motonaga, Color Chart of European Pear `Le Lectier` based on the Color Image,SICE Annual Conference R.Amirulah, M.M.Mokji, Z.Ibrahim, Implementation of Starfruit Maturity Classification Algorithm on Embedded System. 2nd International Conference on Signal Processing Systems (ICSPS), V D. I. Amarasinghe and D. U. J. Sonnadara, Surface colour variation of Papaya fruits with maturity. Proceedings of the Technical Sessions. Xinzhong Wang, Hanping Mao, Xu Han and Jianjun Yin, Vision-Based Judgment Of Tomato Maturity Under Growth Conditions. African Journal of Biotechnology Vol. 10(18), pp Ahmed Jaffar, Roseleena Jaafar, Nursuriati Jamil, Cheng Yee Low, and Bulan Abdullah, Photogrammetric Grading of Oil Palm Fresh Fruit Bunches. International Journal of Mechanical & Mechatronics Engineering IJMME Vol: 9 n. 10. Paolo Gay, Remigio Berruto, Pietro Piccarolo, Fruit Color Assessment for Quality Grading Purposes. ASAE Annual International Meeting/ CIGR XVth World Congress. MPIB, Malaysian Pineapple Industrial Board. [Online]. Available: (Dec. 27, 2010). Mosateru Nogato, Quin Coo Study on Grade Judgment Fruit Vegetables Using Machine Vision. JARQ 32, pp R.Amirulah, M.M.Mokji, Z.Ibrahim, Starfruit Color Maturity Classification Using Cr as Feature, Sixth International Conference on Signal-Image Technology and Internet Based Systems,pp Meftah Salem M. al Fatni, Abdul Rashid Mohamed Sharif, Helmi Zulhidi Mohd Shafri, Osam M.Ben,Omar M. Eshanta, Oil Palm Fruit Bunch Grading System Using Red, Green, and Blue Digital Number, Journal of Applied Science, Zhi-Kai Huang, De-Hui Liu, Segmentation of Color Image Using EM algorithm in HSV, Proceedings of the 2007 I International Conference on Information Acquisition,July 9-11, 2007, Jeju City, Korea Mokji M.M., and Abu Bakar S.A.R, Starfruit Classification Based on Linear Hue Computation. Elektrika Journal of Electrical Engineering, vol.9, n.2, pp Ling Mei Chan, Rodney Tan, Gilbert Thio, Design of Visual base Color Classification System, JASA, January, China Dong Ju Liu, JianYu, Otsu method and K-means,. Ninth International Conference on Hybrid Intelligent Systems,pp Sharmiza Kamaruddin, Sunardi, Kamarul Hawari Ghazali, Bakiss Hiyana, Rina Raha Yellowish Canned Pineapple Scale For Quality Inspection, Cie-TVET, Malaysia. Organized by WorldConferences.net 226