GIAS version 2
Geological Image Analysis Software
How to use it
© Ciaran Beggan and Christopher Hamilton
March 2012
[Link]
Running the application
• In MATLAB
– Add the GIAS directory and sub-directories to your
path (along with the Mcode and skdata directories)
– On the command line type: GIAS_v2
– You should see the GIAS GUI appear within a few
seconds
Panels/Tabs (1)
• Input panel:
– Open files, select processing parameters and options
• Image Analysis Tab:
– Used for image processing statistics
• Nearest Neighbor Analysis Tab:
– Used for nearest neighbor statistics
• Planetary NN Tab:
– Used for analysis of planetary-scale datasets
Panels/Tabs (2) Inputs panel
Output Tabs
Opening an image
• Click on the ‘Load Image’ button
• Select the required image file (tiff, jpeg, png)
Processing the Image
• Select the relevant detection parameters and
output parameters (e.g. pixel size)
• See the Help files for information on the options
• Click on the ‘Process’ button
• It takes less than15 seconds to process test
image LAK_t_23c on an Intel i5 core
Results: Image Analysis Tab
• These are the results for the Image Analysis
Notes on the Image Analysis Tab
• The histograms and graphs are intended for
quickly looking for dominant features only
– The object-size histogram shows the size distribution of the
objects
– The object perimeters shows how circular the objects are
compared to an equal-area circle (e.g. what about crenulations?)
– The eccentricity histograms shows the elongation of the objects
(0 = circular -> 1 = flattened ellipse)
– The rose diagram indicates the direction of the sample fabric
– The information box displays the statistics of the chosen radio
button (e.g. Perimeter length)
• Save the data to file to build and customise your
own figures
Results: Nearest Neighbor Tab
• These are the results for the Nearest Neighbor Tab
Notes of the NN Analysis Tab
• The histograms and graphs are intended for
quick look only
– The NN distance histogram indicates the spread of distances
– The R and c plots give an idea of whether the NN distribution is
compatible with the null hypothesis of Poisson distribution (i.e.
randomly distributed)
– Skew vs. kurtosis gives shows the bias versus peakedness of
the distribution
– If a different model selection is made (e.g Scavanged k=1) the
‘Process’ button must be pressed again to update the figures,
though the NN Results box will be refreshed without rerunning
• Save the data to file to build and customise your
own figures
Saving output (1)
• You can save all the graph calculations using
the ‘Save Results’ button
• This opens a GUI dialogue box with the default
name of ‘Output_’, saved into the current
directory. Change the filename and directory as
required
Saving output (2)
• GIAS writes two Excel-readable tab-delimited files
named from the output filename box: LAK_t_23c_
• It also prints the R and c plots from the NN Analysis Tab
Results in Excel
• You results are
written to two
separate files in
the local
working
directory
• One is for
image analysis
statistics, the
other for
nearest
neighbor
statistics
Results of R and c
• R and c graphs are printed to jpg files, with relevant
statistics and a suggested implication: random, clustered
or uniform
• Again, these are for illustration purposes – use the data
in the NN analysis data file for your own detailed figures
• The data for the bias/error envelopes are stored in the
skdata folder as NNbias_flat_k_0.dat for Poisson test
Other options for NN analysis
• You can load data in as two other types of file
– NN Input image: this is an image with single pixels
representing the centroid or single points (within a
landscape for example). Useful for ArcGIS exports
– NN Centroid List: A list of X and Y coordinates (e.g. in
UTM or another ‘flat’ coordinate system) assumed to
be a two column .txt or .dat file
– Invert Input Image: by default, GIAS assumes that the
objects of interest (e.g vesicles) are the blackest
pixels. Tick this box to invert the image colors (e.g
white becomes black)
Planetary analysis
• This new tab (in version 2) allows analysis of planetary-
scale features or those measured on a spherical object
given in degrees of longitude and latitude
– Analyse NN distributions across the whole sphere, half, third or
quarter sphere.
– Compare to a ‘flat’ Region with size computed or input via the
Hull Area box
• To use this Tab, select the Planetary Data tick box
– An Open File GUI will appear prompting you for projection file,
which gives the radius and flattening of the planet of interest.
– Then load in a two column file consisting of longitude and
latitudes for the points of interest
– The following example is for hotspots on the Jupiter moon Io
Planetary Tab: Load projection file
• Load the ArcGIS projection file (e.g. [Link])
• Then click ‘Load Data’ to load in a two column file of
longitude and latitudes in degrees (e.g. [Link])
Planetary Tab: Load data
• You should see the two columns of data displayed in the Input panel
as a thin line (see upper left). Now hit ‘Process’
Results of hotspots on Io
• The hotspots are randomly distributed when compared against the whole moon’s
area - what about subsets of the data …?
Notes on Planetary Tab
• The Image Analysis and NN Analysis Tabs are ignored once the Planetary
Data Tick box is enabled
– Untick to start using the other two tabs
• Use the ‘[Link]’ projection file if you are analysing items on the
surface of a sphere. Assumes the radius of the sphere is equal to 1.
• The NN distances are also correctly computed with ellipsoidal planetary
projections
• If using the whole sphere, untick the ‘Ignore Hull Boundary Objects’ box.
• Saving the data is similar to the other two tabs. A file with the NN data and
two jpg images of R and c are written out with the
XXX_PlanetNearestNeighborStats.dat ending.
Other options
• See the Help files for advice on how the other options
work
• If the code stops working correctly, the best thing to do is
to close the programme and restart it
• If it is persistently crashing, check the image format is
correct and that you have set the path in Matlab
correctly.
• If all else fails, contact us via the [Link]
website
