Multi-dimensional sampling structure conversion with one-dimensional Nth-band filters

Consideration is given to the design of a specific class of multidimensional filters which are used for sampling structure conversion of multidimensional sequences. Nonrectangular sampling structures are considered. The filter design is based on the concept of generalized separability, i.e. on the use of a cascade of one-dimensional filters operating in the direction of sampling vectors. The author considers the use of one-dimensional Nth-band filters, which result in improved computational efficiency, since the impulse response contains many zero-valued samples. Another advantage of these designs is that, in the interpolation case, the sample values of the input sequence are not changed, which is very natural in image interpolation.<>