Digital Image Analysis for Software Separation and Classification of Touching Grains: I. Disconnect Algorithm

An algorithm to segment connected grain kernel image regions is described. The algorithm attempts to eliminate the need for well-separated kernels in machine-vision-based grain grading experiments. The algorithm uses image transforms from the discipline of mathematical morphology. The algorithm was tested on 746 connected kernels of hard red spring (HRS) wheat, 391 connected kernels of durum wheat, 376 connected kernels of barley, 286 connected kernels of oats, and 393 connected kernels of rye. The spatial arrangement of the touching kernels was random in all of the images used. The algorithm was successful in disconnecting kernels of HRS wheat, durum wheat, barley, oats, and rye with accuracies of 95, 95, 94, 79, and 89%, respectively. The important limitation of the algorithm was that it failed when the connected kernels formed a relatively long isthmus or bridge between them.