0

I am trying to plot 2D surface plots in 3d using mplot3d module.

I know how to "manually" find the max value and its position (in terms of row and column), which is very important in what I am doing. Is there a way to plot that information onto the peak ? That is, writing (maxval,row,col) next to the peak ?

P.S. while I am asking this. Maybe there is an easy way to identify the second peak (or any other peaks for that matter?). I am currently using a mask, to mask out the first peak and to find the second, but I have to be very careful in choosing the sides, because if I happen to mask too little, some of the non-peak stuff will be identified as a peak, and the real second peak will not be identified, messing up a measurement called "peak to peak" signal to noise ratio.

enter image description here Surface Plot

The code I am currently using is :

frame_a = gdal.Open( "frame_{0:05d}.tif".format(274) ).ReadAsArray()
# in case this helps, this is how the images are read, they are 16-bit GS tiffs.    
frame_b = gdal.Open( "frame_{0:05d}.tif".format(287) ).ReadAsArray()


#this does some clever stuff but basically it returns a 2-D 32x32 array.
corr = correlate_windows( windows_a[99], windows_b[99], corr_method = corr_method, nfftx=nfftx, nffty=nffty )

#this is how I find the position of max value.
column = np.argmax(np.max(corr, axis=0))
row = np.argmax(np.max(corr, axis=1))
maximum = corr.max()
print 'column = ' + str(column)
print 'row = ' +str(row)
print 'peak_1 = ' + str(maximum)



import matplotlib.cm as cmps
from mpl_toolkits.mplot3d import Axes3D
from matplotlib.ticker import LinearLocator, FormatStrFormatter

fig = pl.figure()

ax = Axes3D(fig)

# window size is 32 in this case

nx, ny = window_size*2, window_size*2

xx = range(nx)

yy = range(ny)

X, Y = np.meshgrid(xx, yy)

ax.plot_surface(X , Y , corr , rstride = 1, cstride = 1 )

pl.show()
leb
  • 544
  • 1
  • 6
  • 17
  • as a house keeping note, you can't include images until you have a higher reputation and with enough rep you can edit others posts. – tacaswell Jun 26 '13 at 10:59
  • what do you mean by 'loose alignement'? Can you put a second image up some place? You should split the second and third parts of the question into their own questions. The ideal SO thread has exactly one question which makes answering them easier and makes them more useful to future readers. – tacaswell Jun 26 '13 at 11:03
  • Thanks for comments tcaswell ! I edited the question. I think what I am looking for in the other part of the question (now deleted) is actually fixing the "point of view" on the plot (in the above picture, you cannot really see where zeros are). I think this should be easy to find. – leb Jun 26 '13 at 11:58
  • I don't have time to write a real answer, but I think you can do what you want with `annotate` – tacaswell Jun 26 '13 at 12:36

1 Answers1

0

With regards to your second question, you may find the following post useful

Peak-finding algorithm for Python/SciPy

In some work I have done we used a simple approximation of the derivative, when this changes sign you have a peak (in 1D data), one can then add some parameters to remove peaks due to noise. To extend this to 2D I have read about algorithms which follow the gradient (either up or down) until it reaches a maxima. These can be fairly tricky since it is easy to get stuck at local minima/maxima. If you post a follow up to this please link here as I would like to see what people come up with.

Community
  • 1
  • 1
Greg
  • 11,654
  • 3
  • 44
  • 50
  • Ah, of course - calculus is the way ! Now I see why the question should only have one question in it... – leb Jun 27 '13 at 06:39