Decomposition of signals by scale using nonlinear filters

This paper is concerned with a general purpose, multiscale decomposition, namely the datasieve. It is designed specifically to yield multiscale primitives suitable for pattern recognition. The datasieve is appropriate for isolating and locating the position of objects with sharp edges arising from nonlinear events. A typical example is the image due to one object partially occluding another. It can represent structural information in a way that is independent of spatial frequency, has different uncertainty trade-offs, and can he used for scale, position and contrast independent pattern recognition. The datasieve is a nonlinear pyramid in the sense that the signal is simplified at each stage. It is shown that the result of the datasieve decomposition, namely the granularity, can be used directly to recognise patterns in both one-and two-dimensional signals