Abstarct: Image interpolation plays a crucial role in the display, processing, and analysis of medical images. In this research, interpolation methods are classified into three main categories: filter interpolation, ordinary interpolation, and generalized partial volume interpolation. After introducing and analyzing existing methods, a new concept of generalized partial volume interpolation with corresponding constraint conditions is defined, and several new interpolation functions are derived for this category. To evaluate the performance of the methods, experiments on image scaling, rotation, and self-registration were conducted on medical images, and various metrics including entropy, peak signal-to-noise ratio (PSNR), cross-entropy, normalized cross-correlation coefficient, and execution time were measured. Results indicate that among filter interpolation methods, median filter and B-spline filter interpolations perform better. Among ordinary interpolation methods, symmetrical cubic kernel interpolations (especially cubic B-spline) offer higher accuracy, albeit with longer execution times. In the context of image registration, symmetrical generalized partial volume interpolation methods demonstrate superior performance, while asymmetrical methods excel in computational efficiency.