Despite significant advancements in deep learning, effectively capturing fine-grained local details and balancing global context remains a critical challenge in image segmentation, particularly in complex medical scenarios where existing methods struggle with computational efficiency and edge perception. To address these challenges, we propose TransEdge, a novel architecture that integrates a Multi-scale Edge Contrast Sensitivity (MECS) module—a key contribution of this work—into the TransUKAN framework. The MECS module, combining global pooling techniques and multi-scale convolutions, is designed to enhance the model’s ability to perceive edge details and improve local feature extraction, especially in medical image segmentation tasks. Extensive experiments on MNIST, CIFAR-10, and medical image segmentation tasks (BUSI, ISIC, and Kvasir) demonstrate that TransEdge achieves competitive performance compared to state-of-the-art models, with accuracies of 99.16% on MNIST and 92.15% on CIFAR-10. Ablation studies further validate the importance of the MECS module, showing performance improvements on medical image segmentation datasets, with accuracy increases of 1.51%, 0.89%, and 3.63% on BUSI, ISIC, and Kvasir, respectively. Our work provides a lightweight, efficient, and high-performance solution for image segmentation tasks, with significant potential in resource-constrained environments.