IMAGE PROCESSING, FEATURE EXTRACTION TECHNIQUES FOR A POTATO QUALITY SORTING SYSTEM Nicolaas Luwes Image processing is defined as the science or art of taking images and transforming them, enhancing them, highlighting important features and also making images aesthetically pleasing [Elad, A. and Kimmel, R. 2003; Stankovic, L. et al 2000; Nowak, R. D. & Baraniuk, R. G. 1999; Hamza, B and Krim H, 2001; Geusebroek, J. & Smeulders, A. W. M. &van de Weijer, J. 2003; Vertan, C, et al.. 1999]. Image processing is not always concerned with appearances and is applied for measuring level of colour, intensity, extracting quantitative information and many other forms of analysis [Lehmann, T. M. and Palm, C. 2001; Krikke, J. 2005; Chang, C. and Ren, H. 2000; Russ, J.C. 1992; Alippi, C., et al. 2000; Ramboss, B. J. et al.. 2003]. The aim is to extract useful feature information from a potato on a conveyer belt for quality control purposes. Figure 1 Experimental setup The Image possessing setup consists of a turn table with a variety of different quality potatoes on it. The turn table simulates the operation of a conveyor belt system. A camera then take a picture as the potato passes it. The image from the camera is defined as a Linear Time Invariant (LTI) Two-dimensional matrix. [Frey H. Prof. Dr.-Ing. 2005]. Figure 2 Binary image A threshold is applied where certain intensity levels are dissipated to zero and other taken to binary 1, resulting in a binary image [Haralick, R. M. et al 1992, Frey H. Prof. Dr.-Ing. 2005]. Erosion is demonstrated on the binary image with a 3 by 3 matrix and the following equation [Frey H. Prof. Dr.-Ing. 2005, Haralick, R. M. & Shapiro L.G. 1992, Bloomberg, D. S. 2002, Chang, J et al, 1997]: 9 a = e1I e2I e3I e4I e5I e6I e7I e8I e9 = I ei i Figure 3 Noise removed with erosion Eroding the image remove all the unwanted noise as seen in the surrounding areas of Figure 3. Filling the binary image is demonstrated with a 3 by 3 matrix and the following equation [Frey H. Prof. Dr.-Ing. 2005, Haralick, R. M. & Shapiro L.G. 1992, Lee, T. and Lewicki, M, S, 2002]: 9 a = e1U e2U e3U e4U e5U e6U e7U e8U e9 = U ei i Figure 4 holes filled Filling, fills all the holes after which the area is then calculated from the processed binary image. The scale of the irregular shape is calculated with a bonding oval and the area outside of it. Figure 5 bonding oval The colour of the potato is analyzed and illustrated with respect to its area of hue, divided up into red, green and blue [Zhoi, L. & Chalana, V. & Kim, Y. 1998; Naik, S. K. and Murthy C. A. 2003., Deshmukh K.S. and Shinde G. N. 2005]. Figure 6 Hue areas showing the colored areas of the potato The extracted data consisting of potato size, shape and colour, can be fed to a neural network or other decision making software.