Abstract

We describe an effective approach to object or feature detection in point patterns via noise modeling. This is based on use of a redundant or non-pyramidal wavelet transform. Noise modeling is based on a Poisson process. We illustrate this new method with a range of examples. We use the close relationship between image (pixelated) and point representations to achieve the result of a clustering method with constant-time computational cost. Keywords: Cluster analysis, point pattern, `a trous wavelet transform, noise modeling, Poisson distribution, minefield detection. 1 Introduction Point pattern clustering has constituted one of major strands in cluster analysis. We will briefly describe some of this work, in order to motivate the need for (i) a multiscale approach which 1 is computationally very efficient, and (ii) a direct treatment of noise and clutter which leads to improved cluster detection. In the following the first few categories of work proceed in the direction of a m...

Citations