Curve Density Estimates

Abstract

In this work, we present a technique based on kernel density estimation for rendering smooth curves. With this approach, we produce uncluttered and expressive pictures, revealing frequency information about one, or, multiple curves, independent of the level of detail in the data, the zoom level, and the screen resolution. With this technique the visual representation scales seamlessly from an exact line drawing, (for low-frequency/low-complexity curves) to a probability density estimate for more intricate situations. This scale-independence facilitates displays based on non-linear time, enabling high-resolution accuracy of recent values, accompanied by long historical series for context. We demonstrate the functionality of this approach in the context of prediction scenarios and in the context of streaming data.