A new technique for image sample-rate conversion can be efficiently implemented with respect to memory, bandwidth and computational requirements. According to one aspect, the technique takes advantage of certain inherent similarities and symmetries within the image data to limit switching of filters during processing and to otherwise process the data in an efficient order. According to another aspect, the technique can be implemented within a decoding pipeline such as a JPEG/MPEG decoding pipeline in such a way that multiple transfers of image data in and out of the external memory can be avoided. According to a still further aspect, where poly-phase filters are used, computationally-efficient filters are chosen for use with the image conversion process. The technique is amenable to both hardware and software based implementations.