1. INTRODUCTION
Computer vision based quality sorting of apple fruits is necessary for increasing the speed of sorting and eliminating the human error in the process. Research still continues to accurately segment and identify skin defects of apples. For this aim Leemans et al. introduced a Gaussian model of skin color for ‘Golden Delicious’ [1], and a Bayesian classification method for ‘Jonagold’ apples [2], where healthy skin presenting patches was segmented as defected in the former and segmentation of russet defects and color transition areas of skin were problematic in the latter. Rennick et al. used a controlled acquisition system and different classifiers to classify skin color and detect blemishes of ‘Granny Smith’ apples [3]. Yang introduced an automatic system to detect patch-like defects on apples, where he used flooding algorithm to segment defects, structural light and neural networks to find stem-ends and calyxes and snakes algorithm to refine defected area [4]. Unay and Gosselin introduced a neural network based system to segment defects on ‘Jonagold’ apples, where segmentation was accurate, but misclassification of stem-ends, calyxes occurred [5].