Comparison of error bounds for non-parametric dominant point detection

This paper compares three error bounds that can be used to make dominant point detection methods non-parametric. The error bounds are based on the error in slope estimation due to digitization. However, each bound is derived from a different approach. This results into different natures of the three methods and different values. The error bounds can be incorporated in non-parametric framework for dominant point detection. Here, the impact of these error bounds is studied in the context of the non-parametric version of the widely used RDP method of dominant point detection. It is seen that the digital error bound (the third error bound), which depends on both the length and the slope of the line segment, provides the most balanced dominant point detection results for a variety of curves. This analysis is useful for optimal choice of error bound or termination condition in dominant point detection methods.