tive contour model using a variational level set formulation.
By incorporating region-based information into their energy
functional as an additional constraint, their model has much
larger convergence range and flexible initialization. Vemuri
and Chen [9] proposed another variational level set formula-
tion. By incorporating shape-prior information, their model
is able to perform joint image registration and segmentation.
In implementing the traditional level set methods, it is
numerically necessary to keep the evolving level set func-
tion close to a signed distance function [15, 17]. Re-
initialization, a technique for periodically re-initializing the
level set function to a signed distance function during the
evolution, has been extensively used as a numerical rem-
edy for maintaining stable curve evolution and ensuring
usable results. However, as pointed out by Gomes and
Faugeras [12], re-initializing the level set function is ob-
viously a disagreement between the theory of the level set
method and its implementation. Moreover, many proposed
re-initialization schemes have an undesirable side effect of
moving the zero level set away from its original location. It
still remains a serious problem as when and how to apply
the re-initialization [12]. So far, the re-initialization proce-
dure has often been applied in an ad-hoc manner.
In this paper, we present a new variational formulation
that forces the level set function to be close to a signed
distance function, and therefore completely eliminates the
need of the costly re-initialization procedure. Our varia-
tional energy functional consists of an internal energy term
and an external energy term, respectively. The internal en-
ergy term penalizes the deviation of the level set function
from a signed distance function, whereas the external en-
ergy term drives the motion of the zero level set to the de-
sired image features such as object boundaries. The result-
ing evolution of the level set function is the gradient flow
that minimizes the overall energy functional. Due to the in-
ternal energy, the level set function is naturally and automat-
ically kept as an approximate signed distance function dur-
ing the evolution. Therefore, the re-initialization procedure
is completely eliminated. The proposed variational level set
formulation has three main advantages over the traditional
level set formulations. First, a significantly larger time step
can be used for numerically solving the evolution PDE, and
therefore speeds up the curve evolution. Second, the level
set function could be initialized as functions that are com-
putationally more efficient to generate than the signed dis-
tance function. Third, the proposed level set evolution can
be implemented using simple finite difference scheme, in-
stead of complex upwind scheme as in traditional level set
formulations. The proposed algorithm has been applied to
both simulated and real images with promising results. In
particular it appears to perform robustly in the presence of
weak boundaries.
In the following sections, we give necessary background,
describe our method and its implementation, and provide
experimental results that show the overall characteristics
and performance of this method.
2. Background
2.1. Traditional Level Set Methods
In level set formulation of moving fronts (or active con-
tours), the fronts, denoted by C, are represented by the zero
level set C(t)={(x, y) | φ(t, x, y)=0} of a level set func-
tion φ(t, x, y). The evolution equation of the level set func-
tion φ can be written in the following general form:
∂φ
∂t
+ F |φ| =0 (1)
which is called level set equation [11]. The function F
is called the speed function. For image segmentation, the
function F depends on the image data and the level set func-
tion φ.
In traditional level set methods [5–7, 17], the level set
function φ can develop shocks, very sharp and/or flat shape
during the evolution, which makes further computation
highly inaccurate. To avoid these problems, a common nu-
merical scheme is to initialize the function φ as a signed
distance function before the evolution, and then “reshape”
(or “re-initialize”) the function φ to be a signed distance
function periodically during the evolution. Indeed, the re-
initialization process is crucial and cannot be avoided in us-
ing traditional level set methods [4–7].
2.2. Drawbacks Associated with Re-initialization
Re-initialization has been extensively used as a numeri-
cal remedy in traditional level set methods [5–7]. The stan-
dard re-initialization method is to solve the following re-
initialization equation
∂φ
∂t
= sign(φ
0
)(1 −|φ|) (2)
where φ
0
is the function to be re-initialized, and sign(φ)
is the sign function. There has been copious literature on
re-initialization methods [15, 16], and most of them are the
variants of the above PDE-based method. Unfortunately, if
φ
0
is not smooth or φ
0
is much steeper on one side of the
interface than the other, the zero level set of the resulting
function φ can be moved incorrectly from that of the origi-
nal function [4, 15, 17]. Moreover, when the level set func-
tion is far away from a signed distance function, these meth-
ods may not be able to re-initialize the level set function to
a signed distance function. In practice, the evolving level
set function can deviate greatly from its value as signed dis-
tance in a small number of iteration steps, especially when
the time step is not chosen small enough.
So far, re-initialization has been extensively used as a nu-
merical remedy for maintaining stable curve evolution and
Proceedings of the 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR’05)
1063-6919/05 $20.00 © 2005 IEEE