Enhancing Image Segmentation through an Innovative Hybrid Kernel: A Weighted Fusion of Sobel and Prewitt Operators Using Artificial Neural Networks

Image segmentation, a pivotal element in computer vision, relies on effective edge detection to delineate boundaries within images. Traditional methods, including Canny, Sobel, Prewitt, and Laplacian, while widely used, may exhibit limitations that necessitate pre- and post-processing techniques for optimal results. In response to this, our paper introduces an innovative hybrid kernel, constructed from the weighted average of Sobel and Prewitt operators, with weights determined through an Artificial Neural Network (ANN). The objective is to synergize the strengths of these operators, enhancing the precision of edge detection. The proposed hybrid kernel is rigorously evaluated against established techniques, such as Sobel and Canny, utilizing comprehensive metrics like Peak Signal-to-Noise Ratio (PSNR) and F-score. Results consistently demonstrate that the hybrid kernel outperforms previously published outcomes, showcasing notable improvements in edge detection quality and accuracy. This research positions the hybrid kernel approach as a promising alternative for image segmentation, offering advancements beyond conventional methods.