Peridynamic Enabled Fuzzy Logic Edge Detection

Abstract

This dissertation presents a unified framework for robust edge detection and structural featureextraction that integrates the Peridynamic Differential Operator (PDDO) with Fuzzy Logic inference. The proposed approach addresses fundamental limitations of classical gradient-based methods, including sensitivity to noise, dependence on a single threshold, and difficulty in handling discontinuous or low-contrast imagery. The PDDO formulation provides a non-local, mathematically rigorous foundation for computing image gradients with enhanced noise resilience, while the Fuzzy Inference System introduces adaptability and interpretability to classify edge and non-edge regions under uncertain imaging conditions. The framework further employs multithreshold edge-map summation and Hessian-based curvature constraints to preserve weak yet meaningful edges and to differentiate between cracks, blobs, and flat regions. Experimental validation using benchmark datasets for contour detection and the Kaggle Surface Crack Detection dataset for practical defect identification – demonstrates superior performance in terms of edge continuity, localization accuracy, and robustness to noise. The results highlight the framework’s potential for automated inspection, structural health monitoring, and other imaging applications requiring precise and reliable edge detection under challenging conditions.