A Comparative Performance Analysis of Classical and Modern Edge Detection Techniques for Digital Image Processing Applications

Abstract

Edge detection is a core operation in digital image processing and plays a critical role in applications such as image segmentation, feature extraction, object recognition, and computer vision systems. This paper presents a comparative performance analysis of classical and modern edge detection techniques to evaluate their effectiveness under varying image and noise conditions. Classical gradient-based methods, including Sobel and Prewitt operators, and the multi-stage Canny detector are examined alongside hybrid and deep learning–based approaches. A consistent experimental framework is adopted using standard grayscale images, and performance is quantitatively assessed using metrics such as Peak Signal-to-Noise Ratio, Mean Squared Error, edge detection accuracy, and edge continuity. Performance highlights indicate that classical methods achieve acceptable accuracy with minimal computational cost in noise-free conditions but exhibit significant degradation under noise. The Canny detector demonstrates improved robustness and edge continuity due to its noise suppression and hysteresis mechanisms. Modern deep learning–based methods deliver the highest performance, achieving superior accuracy, continuity, and noise resilience, particularly in low-contrast and textured regions, albeit at higher computational expense. Hybrid techniques offer a balanced trade-off, delivering near–deep learning performance with moderate complexity. The results emphasize that optimal edge detection performance depends on application requirements, available computational resources, and real-time constraints rather than accuracy alone.