Edge Detection779 ProMAX® Reference

Edge Detection

The Edge Detection takes the derivative of the data after

plane dip removal. The results highlight dissimilarities in the
data.

Theory
Edge detection takes the derivative of a 9-trace plane after
plane dip removal and assigns that value to the central
sample. In practice, this amounts to weighting the central
sample in relation to the amount of similarity or dissimilarity
in the surrounding traces. If the surrounding traces are
consistently similar, the central sample is given a value near
zero. If the surrounding traces show marked variation, the
central sample is given a non-zero value. Edge detection can
thus give you a clearer image of lateral dissimilarities caused
by such conditions as faulting or stratigraphic changes.

This technique can be applied to conventional seismic data or

to a dataset of similarity attributes. If your input is
conventional seismic data, edge detection preserves
amplitudes but weights them (as opposed to ESP 3D, which
replaces amplitudes with Manhattan distance values). If your
input is 3D ESP data, edge detection accentuates the
differences you uncovered in the Manhattan distance values
generated from the original seismic data.

The first step in Edge Detection is to calculate the local

plane of dip for each sample of the central trace. This is done
in the same way as in the Planar Dip Constrained method of
ESP 3D. Nine values are interpolated where this plane
intersects the center trace and the eight adjacent traces.
These nine values are the input to the edge detection
algorithm, either Laplacian, which uses the second
derivative, or Sobel, which uses the first derivative.

Laplacian Edge Detection

The simpler approach is the Laplacian, which calculates the

second derivative of a 9-trace plane using the relationship
2 2 2 2 2
∇ U = ∂ U ⁄ ∂x + ∂ U ⁄ ∂ y

Other Docs Search Page Known Problems

Edge Detection780 ProMAX® Reference

In practice, the Laplacian algorithm is implemented by

applying a set of sample weights and then combining the
weighted samples. For the 9-trace plane, the center sample is
weighted by +8, and each of the surrounding 8 sample are
weighted by -1.

Sobel Edge Detection

In Sobel edge detection, the first derivative is calculated in

both the inline and crossline directions for a 9-trace plane.
Then a combination of these two orthogonal measurements is
assigned to the center sample.

In practice, the Sobel algorithm is implemented by applying

two sets of sample weights and then combining the weighted
samples. The first weighting mask detects dissimilarity in one
direction (at 90° to the row of zeros):
-1 0 1
-2 0 2
-1 0 1

The second weighting mask detects dissimilarity in the other

direction (at 90° to the row of zeros):
-1 -2 -1
0 0 0
1 2 1

By combining all the weighted samples, dissimilarity can be

detected along any orientation.

The output can be Magnitude, the square root of the sum of

the squares of the two Sobel results, or Phase, the arctangent
of the two results.

Usage
Input data must be sorted with ILINE_NO as the primary sort
key and XLINE_NO as the secondary sort key, or vice versa.
The data must contain at least two ensembles.

Since adjacent traces are used in the calculations, each input

ensemble should contain the same range of crossline or

Other Docs Search Page Known Problems

Edge Detection781 ProMAX® Reference

inline values. Use Pad 3D Stack Volume to pad in any

missing traces within an ensemble.

Parameters

Algorithm

Select the method of edge detection from the following

choices:
• Laplacian performs Laplacian edge detection on a plane
basis. The center sample is weighted by +8, and the sur-
rounding 8 samples are weighted by -1. In a zone of high
similarity, the surrounding samples cancel the central
sample, resulting in near-zero values. In a zone with a
sample dissimilarity, a non-zero value is produced.
• Sobel performs Sobel edge detection on a plane basis.
Two sets of sample weights are applied to detect edges in
two orthogonal directions. Then the weighted samples
are combined using a magnitude or an arc tangent. The
result is a reliable measure of dissimilarities in any
direction.
In a zone of high similarity, the negative weighted
samples cancel the positive weighted samples, resulting
in near-zero values. In a zone with a sample
dissimilarity, a non-zero value is produced.

Sobel component

This appears if Sobel to Algorithm. The options for output

are:
2 2
• Magnitude S 1 + S 2 . This option tends to show large
features in great detail.
• Phase arctan (S2 / S1) in degrees. This option gives a 3D
depth perspective.

Window Length (ms)

Enter the length in ms of the window of data to use in each

search. The window can range from 0 ms (single-sample
multiplication) to the full trace length. Typically you would
use a window of 20 to 100 ms to ensure that adequate data
are included to estimate trace similarity.

Other Docs Search Page Known Problems

Edge Detection782 ProMAX® Reference

On output, one-half Window length plus Maximum dip

search ms of data are zeroed at the beginning and end of
each trace.

Shorter windows give a very sensitive measure of trace

dissimilarity, but they are liable to biasing from noise effects.
Therefore, you should use short windows only with data that
has high signal-to-noise ratios.

Longer windows reduce the risk of noise effects but, because

of the high degree of statistical averaging, may obscure more
subtle features.

Maximum Dip Search (ms/tr)

Enter the maximum geologic dip (in ms/trace) expected to
occur between adjacent traces. Enter 0 if you do not want dip
taken into account in the calculations.

Edge detection is sensitive to dip. We recommend that you

use a nonzero value for this parameter.

The program shifts the semblance windows on neighboring

traces up and down in one-sample increments within the
range specified as it searches for the maximum semblance in
the inline or crossline direction. The purpose is to eliminate
the influence of dip from measurements of trace-to-trace
similarity. For example, if you set this parameter to 12 ms
and have data with a 4 ms sample rate, the program
calculates semblances for time lags of -12 ms, -8 ms, -4 ms,
0 ms, +4 ms, +8 ms, and +12 ms.

Normalization value

Enter a value to subtract from all output trace samples.

The values obtained by edge detection may have either a

positive or negative bias. Applying an appropriate
normalization value shifts the amplitudes so they are fairly
symmetrical around zero and thus can take full advantage of
the SeisWorks® color bar (which can accommodate values
from -128 to +127).

Apply a THRESHOLD to output?

Select Yes to convert all amplitudes to +1 if they are equal to

or greater than the Threshold value or to -1 if they are less
than the Threshold value. The result is a binary output that

Other Docs Search Page Known Problems

Edge Detection783 ProMAX® Reference

clearly differentiates between one range of values and

another.

Applying a threshold can help you distinguish between areas

of like similarity. You can, in effect, separate values that are
close together.

Threshold value

This appears if Yes to Apply a Threshold to output. Enter

the threshold value.

Other Docs Search Page Known Problems

Edge Detection784 ProMAX® Reference

Other Docs Search Page Known Problems