Abstract:
|
Since high dimensional data arises everywhere statistical methods of analyzing high dimensional data is very important. But thinking of the computational cost and the time needed to execute a task, it is also important to study dimension reduction techniques. Statistical shape analysis is a field that often deal with high dimensional data where data analysis is time consuming. Therefore we think dimension reduction is important. In this paper we discuss three different parameterizations that can be used in reducing the number of sampling points of approximating shapes of planar contours. The three parameterizations we discuss are arc length parametrization, which gives equally spaced sampling points that approximates the shape; curvature based parameterization, which chooses points such that total curvature between each consecutive pair of observations is equal; and a third parameterization that is still under study. We use two different decision criteria under each parameterization to select a lower bound for number of sampling points at an error threshold of 0.005. We also discuss how smoothing affects the each of the two decision criteria we use.
|