Multiscale median and morphological filters used for 2D pattern recognition

Rank and morphological filters are closely related. Configured as a multiscale filter bank they yield similar results. The difference between the output of each stage and its input, the granularity, is a complete characterisation of a signal that has parallels with a wavelet characterisation. By matching the granularity of an image with the granularity of a target pattern, a pattern selective system (matched sieve) can be implemented. In some ways it is a nonlinear analogue of a linear matched filter. It is shown, by example, that a matched sieve can outperform a matched linear filter for two dimensional pattern recognition problems