Stm cluster height resizer (stm-cluster-heightzer) is a class that finds peaks of clusters (high bumps) in a scanning tunneling microscope (STM) image(data)
As all the first you need to level the data in the image suitable. I am using Gwyddion for this. So the stm-cluster-heightzer has a method for importing gwyydion data:
gw_file = 'my_folder/my_data.gwy'
all_channels = load_from_gwyddion(gw_file)
Than you get a list like:
[Name: 45-1 Z TraceUp ,
Name: 45-1 Z RetraceUp ,
Name: 45-1 Z TraceDown ,
Name: 45-1 Z RetraceDown ,
Name: Detail 4,
Name: Detail 5]
Choose you data e.g.:
my_pic = all_channels[0]
You can show the data by :
my_pic.show_data()
Now you need to find the peaks in picture
my_pic.find_peaks_in_rows()
my_pic.show_peakXYdata()
Now use
my_pic.group_clusters()
to group thos points to separated clusters
At this stage you can manual add or delet cluster by clickin on it with the mouse. The cluster_peaker will choos the highest point in the near of your click point
ax, picable_artists = my_pic.cluster_peaker()
If you add or deletet some clusters do not forget to update by
my_pic.update_peaked_clusters(picable_artists)
Now you can either cut image with voronoi algorithm
my_pic.cut_image_regions()
ax, picable_artists = my_pic.cluster_peaker(show_regions= True)
Or define some window with pixels as a side of rectangular frame
my_pic.cut_image_regions(window=30)
ax, picable_artists = my_pic.cluster_peaker(show_regions= True, alpha = 0.4)
The Voronoi pattern is more suitable when the clusters are close to each other and it is difficult to separate the clusters with rectangular windows. The Window pattern is probably the most common case and should be used preferably because otherwise the calculation time increases since the larger the windows the more points have to be calculated.
After that the correction of the heights of each cluster can start, the method parralel_correct_height() calculates and corrects the heights by calculating the slopes in each point as described in [1. Horn, B. K. P. Hill shading and the reflectance map. Proc. IEEE 69, 14-47 (1981). ] The script decides what is ground level and what is a peak. Here is only this short explanation, you can read the rest in the docstring.
my_pic.parralel_correct_height(slope_threshold_factor=0.5,
threshold = 'default',
thold_default_factor=0.2,
seek_for_steps='both')
The slope_threshold_factor defines how steep is the slope used to determine the ground level. Must be adjusted mostly on the first image and then it is the same for all similar measurements.
threshold = 'default' means that the criterion for the separation of individual steps in ground level is taken as the square root of the sum of the distances between the maxima and minima coordinates sqrt( abs(x_max -x_min)^2 + abs(...) ) (see parameter t for scipy.cluster.hierarchy.fcluster(Z,t)). With thold_default_factor you can scale this parameter 0.9 means 90%
If the parameter seek_for_steps = 'both' is set then the script tries to determine the steps in the region in addition to the averaged value of the ground level. It tries to group the points with the parameter thold_default_factor. The larger the factor, the further apart the centers of the grouped points may lie (see scipy.cluster.hierarchy.fcluster).
as you can see, 2 groups were found in the ground level. The middle point (midpoint between 4 points: maximum and minimum x/y coordinates) of the "Groupe:0" is closer to the coordinates of the peak only 1.12 pixel. So this is taken as the ground level belonging to the peak and the average of these points is subtracted from the original height of the peak and stored in the instance true_hight_closest_ground_level.
The adjustment can also be performed on individual regions, if needed. Hier just an example with wrong threshold
my_pic.regions[95].thold_default_factor=0.2
__= my_pic.regions[95].find_groundlevel()
__= my_pic.regions[95].calc_true_hight()
my_pic.update_height(my_pic.regions[95],95) #you need to update the list of results
my_pic.regions[95].plot_ground_level(axis_view =(0,0))
An finaly you get an PandasDataFrame with all cluster coordinates, heights and corrected heights.
my_pic.heights
The x,y coordinates are in pixels and x_m/y_m coordinates are in physical dimensions (in this case meters). The coordinates in pixels are used for simplicity, because so the search for pixels with the eye, in my opinion, is easy.
By using plot_heights_distribution() you can see some statistics (plotly interactiv image):
my_pic.plot_heights_distribution()
