Implementation Of Morphological Skeletonization For Leaf Vein Extraction From Leaf Images

Abstract

This study implements the Morphological Skeletonization method to automatically extract leaf vein structures from mango leaf (Mangifera indica) images. The developed system utilizes an image processing pipeline that includes HSV-based leaf segmentation, preprocessing with Bilateral Filter, vein contrast enhancement using dual-scale Black-Hat Transform and Contrast Limited Adaptive Histogram Equalization (CLAHE), binarization using Otsu and Adaptive Thresholding methods, morphological opening and closing operations, and skeletonization using the iterative thinning algorithm from the scikit-image library. Performance evaluation was conducted quantitatively using Mean Squared Error (MSE) metrics at two stages, namely Binarization MSE and Skeleton MSE. Testing was performed on a dataset consisting of 150 mango leaf images in JPG format collected using a smartphone camera under standardized acquisition conditions. The results show that the system achieves an average Skeleton accuracy of 94.47% with a standard deviation of 1.06%, an average Binary accuracy of 61.28%, and an average Global accuracy of 77.87%. The average Binarization MSE of 25179.25 and Skeleton MSE of 3599.02 confirm that the system is capable of extracting leaf vein skeletons consistently and accurately across the entire dataset. The extracted skeleton structural features include length, number of branch points, connectivity, and density, which quantitatively represent the morphological characteristics of the leaf venation network. This study demonstrates that the mathematical morphology approach offers an efficient, transparent, and interpretable solution for leaf vein extraction tasks compared to deep learning-based approaches that require large computational resources.