Detection of defects on fruits by machine vision and unsupervised segmentation

Defect detection on fruits by machine vision is a complex task. Indeed, the sound tissue colour is not uniform and the defects present a wide variability in colour, shape and texture. Mostly often, images are acquired by conventional RGB cameras and defect segmentation is performed by algorithms based on Bayes' rules. The efficiency of these methods can be improved firstly by acquiring images with a dedicated vision system (multi-spectral imager) and secondly by implementing unsupervised segmentation methods (method derived from the 'mean shift' procedure of Camaniciu and Meer, 2002).