Region growing Euclidean distance transforms

Abstract

By propagating a vector for each pixel, we show that nearly Euclidean distance maps can be produced quickly by a region growing algorithm using hierarchical queues. Properties of the propagation scheme are used to detect potentially erroneous pixels and correct them by using larger neighbourhoods, without significantly affecting the computation time. Thus, Euclidean distance maps are produced in a time comparable to its commonly used chamfer approximations.