The PMN algorithm

The "Euclidean Distance Transformation by Propagation using Multiple Neighborhoods", or EDT-PMN works as follows:

Algorithm 3   EDT by Propagation with Multiple Neighborhoods  

In this algorithm, pixels p that do not propagate in the first propagation loop remain in the bucket structure. This is accomplished by storing them in a temporary buffer list, that becomes bucket(d) after all pixels in bucket(d) were processed. During the second propagation, the size of the neighborhoods $N_{\min}(d)$ increases as the right-most column of table . For $2 \leq d < 116$, $N_{\min}(d)$ is the $3 \times 3$ neighborhood. For $116 \leq d < 520$, it is the $5 \times 5$ neighborhood, and so on.