Comparative Evaluation of Image Interpolation Techniques for Lung Image Resizing in Medical Applications

In medical image processing, resizing images is an essential step for preparing a consistent image size and also for preparing the dataset for training various application models, including computer vision, image processing, machine learning, and diagnostic imaging. It can change the number of pixels in the image, which affects both the visual quality and the performance of the models. In this paper, we evaluated five interpolation techniques, Area, Bicubic, Bilinear, Lanczos, and Nearest Neighbor, used for image resizing and experimented with 933 lung X-ray input medical images. Structural Similarity Index (SSIM), Mean Squared Error (MSE), and computation time are measured to evaluate the performance of the resized images. The experiment with the lung image dataset shows that the Bilinear Interpolation technique achieves the lowest MSE value of 0.002 and the highest SSIM value of 0.999, making it the most suitable resizing technique for medical images. The Lanczos and Bicubic techniques offer lower quality and better balance of accuracy and processing time compared to the Area Interpolation Technique, which produces a poor MSE value. Nearest Neighbor interpolation is the fastest method, but it generates poor image quality and is unsuitable for medical applications. The results highlight the right interpolation technique based on the quality and computational constraints of the specific image processing tasks.