Plant Cell Segmentation with Adaptive Thresholding

There are many approaches to plant cell segmentation, but there is no established method to segment plant cell for a portable, USB-powered optical microscope. Existing methods leverage on sophisticated microscope such as confocal laser scanning microscope or electron microscope may not be applicable for a portable setup. Staining of plant cell specimens, in order to improve visibility of boundaries, might affect the plant cell and also requires additional preparation work prior to acquisition which could be infeasible for on-the-fly applications. Conventional plant cell segmentation using watershed transform often results in over-segmentation, hindering the effectiveness of the method. Hence, we propose a thresholding method based on Otsu's method, to retain majority of the image information to improve the success rate of the cell segmentation. The method is implemented on a leaf cellular image acquired from freshwater weed elodea. The region identified by the improved watershed transform can be further processed to locate the centroids of the cells. We experimented our method on images filled fully with plant cells and filled partially with plant cells. We also studied the impact of boundary definition of the image to our method.