A time-frequency map can also be plotted as a function z(f,t), as in this graph, where z is the value of the power at a given time and frequency. A curve in the map becomes a "ridge" in the function.	It is difficult to find a ridge of unkown shape. However, we can make the shapes of ridges more uniform by convolving them with a two-dimensional Gaussian distribution. Such a distribution is shown in this graph.
The result of the convolution is shown in this graph. The ridge in the graph is characterized by two facts: 1) the second derivative perpendicular to the ridge is highly negative along the peak of the ridge. 2) the first derivative perpendicular to the ridge vanishes along the peak of the ridge.	To find the direction in which the second derivative of a function S(t,f) is maximum for each point (t,f), we construct the Hessian matrix at each point. One of the eigenvectors of this matrix points in that direction. We then look for points on the graph where the second derivative in that direction has a high magnitude and the first derivative vanishes.