Research on conjoint grains of rice based on machine vision

In this paper, a new watershed algorithm based on prior knowledge is presented for the recognition of binary images of conjoint grains of rice by machine vision. First, an algorithm for component labeling is used to find whole components of the conjoint grains of rice in each binary image. After the components of all the single grains are removed from each binary image according to their area threshold (AT), only the components of the conjoint grains remain, which are marked by M_S. The remaining components of the conjoint grains are then separated from each other by the watershed algorithm where the threshold (E_T) of the efficient erosion times is set, and the corresponding segmented components are marked by M_C. Finally, the union of M_S and M_C is considered as the result of recognition of binary images with conjoint grains. Over 95% of connected rice particles are correctly segmented, which indicates that the proposed method is valid and efficient.