An Autoregressive Curvature Model for Describing Cartographic Boundaries

This paper introduces a new method for representing cartographic boundaries using autoregressive model parameters. An autoregressive model is presented to model a time series which describes the correlated shape variations determining the overall shape of the boundary. Such time series are obtained using an appropriate nonperiodic sampling sequence from a curvature function of the boundary. In the nonperiodic sampling scheme the sampling points are a set of highly informative and significant points. To reduce the sensitivity of curvature to the noise in boundary, the curvature values are estimated at each point using a filter with a proper degree of smoothing. The univariate AR model parameters are invariant to translation, scale, and rotation of the boundary, and as a consequence this modeling can be used to produce invariant recognition and description of cartographic boundaries.