how much of the available dynamic range is used by an image
(Gonzalez and Woods, 2008).
The dynamic range is represented by how many grey levels in
the grey-scale are occupied. For instance, the pixel values of
Figure 1-a fall between grey values 0 and 220 (within a range of
0 to 255) with none in the other regions. This indicates a
relatively wide dynamic range of brightness (Cunha et al. 2006)
thus, in general, requiring a larger contour representation (hence
nodes) than an image with a narrow grey-scale distribution
(small dynamic range).
2. GENERATING 3D DATA POINTS
The process of generating 3D data points from contours
comprises the step of assigning coordinate values to pixels in
the image. This is followed by interpolating between pixels to
find the coordinates of points in the path of a contour having the
same colour intensity value.
For instance, Figure 1(a) shows the original image of Peter. tif
(400%) whereas F igure 1(b) is the results of joining
corresponding contour nodes using line segments. By way of
example, the contour increments in Figure 1(b) is 8, ranging
between 0 and 220 grey-scale values or colours. In this instance,
bilinear interpolation was used to determine the contour nodes
(Watson, 1992).
Figure 1. (a) The original image of Peter.tif (400°), whereas (b)
is a contour representation of (a) for contour increments equal to
8 grey-scale values.
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B3, 2012
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia
In the interest of reducing to a minimum the number of nodes or
contour vertices, and thus minimise data Storage and
transmission time, it is often possible to represent the contours
using fewer line segments. It is important however, to maintain
sufficient details so as to portray important features such as
minute wavers and concavities. In most real world Cases, an
image will contain contour portions with little direction changes
(i.e. straight regions) and portions with a lot of direction change
(i.e. curved regions) as shown in Figure 2(a).
A variation of the basic contouring approach, referred to as
variable line segment boundary representation (Baxes, 1994)
can be used to better represent small contour segments changes
without using small line segments throughout the contour
description.
Therefore, this technique can be used to remove nodes from
contours as it combines multiple line segments and replaces
them with a single longer segment over portions of the contour
that have small direction changes. A variable number of contour
nodes are covered by a single line segment, depending on how
contour direction occurs. The nodes that are eliminated during
this process may be recovered if required in the reconstruction
process.
This way, the line segments of a given contour will not unduly
record multiple short line segments (and thereby nodes) where
the contour direction change only a little. The implementation
of the line segment approach is carried out by first performing a
line segment contouring operation, using line segment lengths
that are sufficiently short so as to contain the contour details
required to reconstruct an original raster image.
Then, the angles of each pair of adjoining line segments are
compared. Where the directional angles are similar or identical,
the two lines segments are combined to create a single line
segment representing their combined length and overall angle
(see Figure 2(b)). The end result is a contour segment
representation that can be significantly more concise, while
maintaining a high level of accuracy.
*
® a € *
LÀ a a
» + a
* *
. « *
À
+ e
e .v*
. * e * o. €
= «
æ ® Li
* €
. e
* .* +
? e
* ®
Figure 2. (a) A contour often contain regions of little change
and regions of great change and (b) The variable line segment
description combines adjoining contour segments with similar
direction angles (after Baxes, 1994)
Another technique which may assist in removing redundant
nodes from the reconstruction process is to compare the area
these contours occupy and the shorter and longest distance to
their centroids (Bonham-Carter, 2006). These parameters may
offer the best statistical constraint for deciding whether a
contour (and its nodes) is very similar to another and thereby
could be removed.
Also, within any group of at least three close nested contours it
may possible to test for redundant nodes by removing the
mid
two
smc
cha
nec
ima
For
link
(i.e
con
def
ver
In
on
req
pix
reg
int
acc
Th
Es
ref
val
for
int
the
