In [0]:

    

!git clone https://github.com/aplyer/gefolki.git

import sys
sys.path.append('gefolki/python')
from algorithm import EFolki


    

In [0]:

    

!wget https://raw.githubusercontent.com/aplyer/gefolki/master/datasets/JacksonvilleNavalAirStation_sandiaKu.png
!wget https://raw.githubusercontent.com/aplyer/gefolki/master/datasets/S1_Jacksonville_GEE.tif
#  'JacksonvilleNavalAirStation_sandiaKu is Courtesy of Sandia National Laboratories, Radar ISR'


    

In [47]:

    

import numpy as np
from skimage.io import imread
import pylab as pl

M = imread("S1_Jacksonville_GEE.tif")
pl.figure()
pl.imshow(M)
pl.title('Master')
pl.show()

  
S = imread("JacksonvilleNavalAirStation_sandiaKu.png")
pl.figure()
pl.imshow(S)
pl.title('Slave: Image to find within the Master')  
pl.show()


    

In [0]:

    

Master=M[:,:,0];
Slave=S[:,:,0];
dimx,dimy=np.shape(Master);
dimxn,dimyn=np.shape(Slave);


    

In [0]:

    

fdecimation = 8
rank=4

# Decimation
from skimage.transform import resize
nx = int(round(dimx/fdecimation)) 
ny = int(round(dimy/fdecimation))
Mg = resize(Master,(nx, ny),1,'constant')
nsx = int(round(dimxn/fdecimation)) 
nsy = int(round(dimyn/fdecimation))
Sg = resize(Slave,(nsx, nsy),1,'constant')       

# Rank computation and Criterion on images after deximation
from rank import rank_sup as rank_filter_sup
from rank import rank_inf as rank_filter_inf
Mg_rank = rank_filter_inf(Mg, rank) # rank sup : high value pixels have low rank
Sg_rank = rank_filter_inf(Sg, rank)
                 
R=np.zeros((nx-nsx-1,ny-nsy-1)); 
indices=np.nonzero(Sg_rank);
test2=Sg_rank[indices];

for k in range(0,nx-nsx-1):
    for p in range(0,ny-nsy-1):
        test1=Mg_rank[k:k+nsx,p:p+nsy];
        test1=test1[indices];
        test=(test1-test2)**2
        R[k,p]=test.mean();  
        
pl.figure()
pl.imshow(R,vmin=np.min(R[np.nonzero(R)]),vmax=R.max(),cmap='jet')
pl.title('Criterion after decimation')  
pl.show()
  
ind = np.unravel_index(np.argmin(R, axis=None), R.shape)
indx=(ind[0]*fdecimation)
indy=(ind[1]*fdecimation)


# Rank computation and Criterion on images without deximation after preinitialization
Master_crop = Master[indx-100:indx+dimxn+100,indy-100:indy+dimyn+100]
M_rank_crop = rank_filter_inf(Master_crop, rank) # rank sup : high value pixels have low rank
S_rank = rank_filter_inf(Slave, rank)

dimxcrop,dimycrop=np.shape(M_rank_crop);
Rfin=np.zeros((dimxcrop-dimxn-1,dimycrop-dimyn-1)); 
indices=np.nonzero(S_rank);
test2=S_rank[indices];

for k in range(0,dimxcrop-dimxn-1):
    for p in range(0,dimycrop-dimyn-1):
        test1=M_rank_crop[k:k+dimxn,p:p+dimyn];
        test1=test1[indices];
        test=(test1-test2)**2
        Rfin[k,p]=test.mean();
        
# Final Extraction
ind = np.unravel_index(np.argmin(Rfin, axis=None), Rfin.shape)
indx=(ind[0])
indy=(ind[1])
Master_final = Master_crop[indx:indx+dimxn,indy:indy+dimyn];

pl.figure()
pl.imshow(Rfin,vmin=np.min(Rfin[np.nonzero(Rfin)]),vmax=Rfin.max(),cmap='jet')
pl.title('Fine Criterion without decimation after préinitialisation')  
pl.show()


rgb = np.dstack((Slave,Master_final,Slave))

pl.figure()
pl.imshow(rgb)
pl.title('Superposition of Slave and Master Extraction')
pl.show()