The development of artificial intelligence (AI) and image processing could help streamline the examination of the skin lesions, reducing subjectivity and avoiding the painful and unnecessary biopsies. The image resizing is an image pre-processing step usually employed to fit the input of the AI networks and accelerate the learning process of deep learning models. However, it could affect the inner characteristics of the image, resulting in low-quality images and coarse segmentation of the skin lesions. This paper examines how image resolution and interpolation techniques affect U-net-based skin lesion segmentation. Bi-Cubic gave better results for higher resolutions but at higher cost. The extensive experiments demonstrated better image quality using MSE, PSNR, SSIM, and FSIM when resizing to higher image resolutions using Bi-Linear, Bi-Cubic, and Nearest Neighbor interpolation techniques. Moreover, Nearest Neighbor and Bi-Linear techniques allow for a better trade-off between the segmentation performance and the training time. On the other hand, the bi-cubic technique combined with a training resolution of 512 × 512 has offered the highest performance of JI = 90,80% and ACC = 94,97% but at the price of increased computational cost. Surprisingly, lower-quality Nearest Neighbor images minimally impact segmentation, aiding efficient model optimization for medical imaging.