Automatic defect segmentation of 'Jonagold' apples on multi-spectral images: A comparative study

Several thresholding and classification-based techniques were employed for pixel-wise segmentation of surface defects of 'Jonagold' apples. Segmentation by supervised classifiers was the most accurate, and the average of class-specific recognition errors was more reliable than error measures based on defect size or global recognition. Segmentation accuracy improved when pixels were represented as a neighbourhood. The effect of down-sampling on segmentation accuracy and computation times showed that multi-layer perceptrons were the best. Russet was the most difficult defect to segment, and flesh damage the least. The proposed method was much more precise on healthy fruit. (c) 2006 Elsevier B.V. All rights reserved.