Segmentation of touching seeds based on shape feature and multiple concave point detection

For more accurate computing result of touching vegetable seeds, a novel improved segmentation method combining morphological features and corner detection based on curvature scale space is proposed. The original images taken by the industrial camera are preprocessed by de-noise, enhancement, and contour extraction operations. Then, the corner detection based on curvature scale space is conducted for the obtained contours. In this part, a quick and simple method for distinguishing the concave-convex quality of contour corners based on area characteristics is introduced, and new matching rules are proposed to improve concave point matching algorithm. Finally, the touching vegetable seed kernels are separated by connecting matched hollow points by adopting Bresenham algorithm. Shape features (mainly includes perimeter and area) are extracted from each segmented area and classified to determine whether each separate area is a single seed. If not, previous segmentation steps will be repeated until each connected domain has the similar features. The algorithm was tested for different types of vegetable seeds with the various touching degree. Compared to the existing algorithm, this algorithm eliminates the traditional limitations that seeds need to be placed in a nontouching manner or all adherent ones need to be put in the same layer during imaging process. The algorithm achieves effective segmentation for complicated scenarios of the touching or partial overlapping vegetable seeds.