Self-organizing map combined with a fuzzy clustering for color image segmentation of edible beans

A novel segmentation approach that partitions color images into two uniform regions is described. This unsupervised procedure is based on a self–organizing map neural network and fuzzy c–means clustering (SOM_FCM). The self–organizing map allows the mapping of a color image related to edible beans into a consistent two–dimensional table through a non–linear projection. Fuzzy clustering is then applied to the Kohonen map to determine the two cluster centers. The results were compared to a standard spatial thresholding segmentation method. The two segmentation approaches were used for the segmentation of 150 color images of beans (acceptable, small, damaged, and broken), foreign materials, and stones. The results showed that the SOM_FCM outperformed the spatial thresholding method in identifying objects. It was found that the size of the Kohonen layer, the form of the neighborhood function, and the mapping topology did not have a significant effect on the segmentation performance of the SOM_FCM. The average percentage of correctly matched pixels was 99.31% for the SOM_FCM and only 89.71% for the spatial thresholding method. Unlike the SOM_FCM, the spatial thresholding method failed to correctly segment most of the broken bean and stone images. Unsupervised neural networks have the potential to improve agricultural machine vision applications.