Mathematical Morphology
-
Set theoretic, algebraic, mathematical morphology
operations can be viewed as geometric deformations described by curve evolutions
governed by a partial differential equation. While typical curve evolution
implementations rely on embedding the curve as the zero level set of an
evolving surface, this approach cannot handle open curves and junctions,
and is inefficient due to the additional dimension. We have presented a
novel approach relying on an analytic wave propagation framework, but one
which is implemented on a discrete grid and using discrete directions,
by relying on subpixel geometric models. The resulting implementation
is exact for the class of piecewise circular curves and its usefulness
is demonstrated for extracting skeletons and for smoothing open curves
and shapes.
-
We propose a wave propagation algorithm based on
the intermediate view of Huygens' principle and
Fermat's principles. In the Huygens' wave
propagation construction, each point on the wavefront emanates light
rays in the form of a one parameter family of radial curves. This can be
applied to discrete domain rather easily, but it is inefficient due to
excessive overlap. In Fermat's principle, the light rays propagate orthogonal
to the wavefront and the new wavefront may be constructed by moving along
the light rays. However, tracing the light rays on the discrete domain
would leave gaps on the wavefront. Instead of using a full circle in the
first case and a thin beam in the second, we proposeusing a sector of discrete
beams with minimal overlap as in the CEDT algorithm [Ragnemal:neighborhoods:cviu92],Specifically,
the CEDT algorithm which typically uses discrete wave directions
from discrete boundary locations is now generalized to free-form
curve segments,yet maintaining propagation using discrete grid locations
and directions. In our approach, instead of propagating only distance information
(time), we also propagate geometric models of the sub-pixel boundary.
Analytic distance values from the source
boundary segments are then computed at each step of propagation process.
-
Wavefront is represented by a set of discrete points
-
Each discrete point on a wavefront contains
-
direction
-
distance
-
source
-
Propagation takes place from the smallest value
-
Analytic distance values from the source boundary segments are then computed
at each step of propagation process
-
Propagation of discrete waves from a
geometric boundary model
-
Examples of morphological operations via discrete wavefront propagation
algorithm:
-
An example of the opening operation using subpixel evolution (CLICK ON
THE IMAGES FOR HIGHER RESOLUTION!)
-
Smoothing open curves by subpixel closing (CLICK ON THE IMAGES FOR
HIGHER RESOLUTION!)