Segmentation Methods of Fruit Image and Comparative Experiments

Abstract

Fruit image segmentation issue on color difference between mature fruits and backgrounds under natural illumination condition is an important and difficult content of fruit-harvesting robot vision. Some studies concerning fruit image segmentation have been presented in the last few years. However, these studies are focused on particular fruit and different from segmentation results. In this paper, four kinds of segmentation methods are presented and applied into fruit image segmentation. The tests show that these methods can segment successful several kinds of fruits image, such as apple, tomato, strawberry, persimmon and orange. Dynamic threshold segmentation method has better performance and least cost time than extended Otsu method, improved Otsu combined with genetic arithmetic and adaptive segmentation method based on LVQ network. Meanwhile, it has satisfactory effect upon fruit image under natural illumination condition. Adaptive segmentation method based on LVQ network can only be applied into balanced color instance of particular fruit, and it isnpsilat adapt to be applied into real-time occasion because of high cost time.