Plant petioles and stems are hierarchical structures comprising cellular tissues in one or more intermediate hierarchies displaying quasi random to heterogeneous cellularity that governs the overall structural properties. Exact replication of natural cellular tissue leads to the investigation of mechanical properties at the microstructural level. However, the micrographs often display artifacts due to experimental procedure and prevent representative spatial modeling of the tissues. Existing methods such as local thresholding or global thresholding (Otsu’s method) fail to effectively remove the artifacts. Hence, an efficient algorithm is required that can effectively help to reconstruct the geometric models of tissue microstructures by removing the noise. In this work, perception-based thresholding that conceptually works like human brain in differentiating noise from the actual ones based on color is introduced to remove discrete (within a cell) or adjacent (to the cell boundaries) noise. A variety of image dataset of non-woody plant tissues were tested with the algorithm, and its effectiveness in eliminating noise was quantitatively compared with existing noise removal techniques by Bivariate Similarity Index. The bivariate metrics indicate an enhanced performance of the perception-based thresholding over other considered algorithms.
An a-contrario approach to quasi-periodic noise removal