Journal of Mathematical Imaging and Vision
https://doi.org/10.1007/s10851-018-0805-1
Distance Functions Based on Multiple Types of Weighted Steps
Combined with Neighborhood Sequences
Benedek Nagy
1
· Robin Strand
2
· Nicolas Normand
3
Received: 7 June 2017 / Accepted: 3 March 2018
© Springer Science+Business Media, LLC, part of Springer Nature 2018
Abstract
In this paper, we present a general framework for digital distance functions, defined as minimal cost paths, on the square grid.
Each path is a sequence of pixels, where any two consecutive pixels are adjacent and associated with a weight. The allowed
weights between any two adjacent pixels along a path are given by a weight sequence, which can hold an arbitrary number of
weights. We build on our previous results, where only two or three unique weights are considered, and present a framework
that allows any number of weights. We show that the rotational dependency can be very low when as few as three or four
unique weights are used. Moreover, by using n weights, the Euclidean distance can be perfectly obtained on the perimeter of
a square with side length 2n. A sufficient condition for weight sequences to provide metrics is proven.
Keywords Distance functions · Weight sequences · Neighborhood sequences · Chamfer distances · Approximation of
Euclidean distance
1 Introduction
In a digital space (given for example by the pixels on a com-
puter screen), not all properties of the Euclidean geometry
are fulfilled. This is mainly due to the discrete (as opposed to
continuous) structure of digital spaces. In the digital geom-
etry framework, a classical way to define distance functions
and metrics is by means of shortest paths or minimal cost
paths. By doing so, the distance between two points is sim-
ply the sum of weights along a given path. With appropriate,
efficient algorithms, this allows completely error-free, and
fast, computation of the distances in two and higher- dimen-
sional digital spaces [1,17,21].
B Benedek Nagy
nbenedek.inf@gmail.com
Robin Strand
robin@cb.uu.se
Nicolas Normand
Nicolas.Normand@polytech.univ-nantes.fr
1
Eastern Mediterranean University, Famagusta, North Cyprus
Mersin-10, Turkey
2
Centre for Image Analysis, Uppsala University, Uppsala,
Sweden
3
LS2N UMR CNRS 6004, Université de Nantes, Nantes,
France
The digital approach we follow, where connected paths are
used to define distance functions, is fundamentally different
from the approach when Euclidean distances are computed.
This is usually done by vector propagation [5,14], fast march-
ing [16], or by separable algorithms [2,4]. These algorithms
either result in errors (i.e., deviations from the metrics under
consideration due to approximation errors or deficiencies in
the algorithm definition) and/or they don’t generalize to so-
called constrained distance transform. See the discussion in
[18]. Therefore, we believe that it is important to develop
both the theory based on these digital distances and practical
algorithms for image processing that can effectively utilize
these distances.
The cityblock and chessboard distances are the two dig-
ital distances first described in the literature [15]. These
distance functions have high rotational dependency, but are
easy and efficient to compute. The theory of digital distances
has developed rapidly from the 1980s. The weighted (cham-
fer) distances [1], where the grid points together with costs
to local neighbors form a graph in which the minimal cost
path is the distance, are a well-known and often used con-
cept in image processing. Contrary to weighted distances, the
allowed steps may vary along the path with distances based
on neighborhood sequences from a predefined set of steps [6],
for instance, by mixing the cityblock and chessboard neigh-
borhood. In [20,22], the concept of weighted distances is
123