96 lines
4.1 KiB
Python
96 lines
4.1 KiB
Python
'''deconvolving 2D images'''
|
|
|
|
import numpy as np
|
|
import h5py
|
|
import argparse
|
|
from scipy.signal import convolve
|
|
from scipy.ndimage import gaussian_filter
|
|
from deconvolve_func import DeconvolveWithBead as DWB
|
|
|
|
parser = argparse.ArgumentParser()
|
|
parser.add_argument('bead', type = str, help = "original file")
|
|
parser.add_argument('--output', type = str, required = True, help = "file name for the output of images")
|
|
parser.add_argument('-pd', '--plot_data', type = str, required = True, help = "data storage file name, iterations added automatically")
|
|
parser.add_argument('-it', '--iterations', type = int, required = True, help = "nr of iterations")
|
|
parser.add_argument('-in', '--intensity', type = float, nargs = '+', required = True, help = "image intensity; division of the signal by intensity")
|
|
parser.add_argument('-k', '--kernel', type = float, nargs = '+', required = True, help = "kernel g sigma values")
|
|
parser.add_argument('--vox', type = float, nargs = '+', default = [1.0, 1.0, 1.0], help = "voxel values in micrometers, default [1.0, 1.0, 1.0]")
|
|
|
|
args = parser.parse_args()
|
|
|
|
|
|
#--------- importing the image, that will become the GROUND TRUTH ------
|
|
with h5py.File(args.bead, 'r') as hdf5_file:
|
|
original = hdf5_file['t0/channel0'][:]
|
|
|
|
print("algse pildi integraal", np.sum(original))
|
|
|
|
#--------- creating the 2D gaussian PSF----------------------------------
|
|
points = np.max(original)
|
|
point_sources = np.zeros(original.shape)
|
|
|
|
center_x = original.shape[0] // 2
|
|
center_y = original.shape[1] // 2
|
|
|
|
point_sources[center_x, center_y] = points
|
|
#point_sources[90, 110] = points
|
|
psf = gaussian_filter(point_sources, sigma = 2.4)
|
|
|
|
print("psf integraal", np.sum(psf))
|
|
if not np.isclose(np.sum(psf), 1.0, atol=1e-6):
|
|
psf /= np.sum(psf) # normaliseerin, et piksli väärtused oleksid samas suurusjärgus
|
|
print("psf-integraal-uuesti", np.sum(psf))
|
|
|
|
#-------------- DECONVOLUTION--------------------------------------------------------------
|
|
with h5py.File(args.output, 'w') as hdf5_file:
|
|
'erinevad mürad ### signal intensity #################################################'
|
|
intensity = args.intensity
|
|
|
|
for i in intensity:
|
|
scaled_original = original / i if i!=0 else original.copy()
|
|
scaled_original = scaled_original.astype(np.float64)
|
|
image = gaussian_filter(scaled_original, sigma = 2.4)#convolve(scaled_original, psf, mode='same')
|
|
image[image <= 0] = 0
|
|
|
|
"rakendan pildile Poissoni müra #################################################"
|
|
if i == 0:
|
|
noisy_image = image.copy()
|
|
else:
|
|
noisy_image = np.random.poisson(lam = image, size = image.shape).astype(np.float64)
|
|
|
|
hdf5_file.create_dataset(f"noise_level_{i:08.3f}", data = noisy_image)
|
|
hdf5_file.create_dataset(f"scaled_original_{i:08.3f}", data = scaled_original)
|
|
|
|
'''kernelid erinevate sigmadega - pildi taastamine ##################'''
|
|
g_sigma = args.kernel
|
|
vox = np.array(args.vox)
|
|
mses = []
|
|
mssim = []
|
|
|
|
for j in g_sigma:
|
|
image_kerneled = gaussian_filter(noisy_image, sigma = j)
|
|
psf_kerneled = gaussian_filter(psf, sigma = j)
|
|
print(np.sum(image_kerneled),np.sum(noisy_image), np.sum(psf_kerneled), np.sum(psf))
|
|
'dekonvolveerin - taastan pilti ######################################'
|
|
deconv = DWB(image = image_kerneled, voxel_size = vox, bead = scaled_original)
|
|
|
|
dec = deconv.deconvolve_rl(n_iterations = args.iterations,
|
|
bead = scaled_original,
|
|
psf = psf_kerneled,
|
|
im = image_kerneled)
|
|
|
|
hdf5_file.create_dataset(f'intensity_{i:08.3f}/SIGMA_{j:03.1f}', data = dec[0])
|
|
hdf5_file.create_dataset(f"intensity_{i:08.3f}/minimum_image_{j:03.1f}", data = dec[4])
|
|
|
|
mses.append(dec[1])
|
|
mssim.append(dec[2])
|
|
|
|
|
|
print("---------------lõpp-pildi integraal", np.sum(dec[0]))
|
|
data = np.column_stack((np.column_stack(mses),np.column_stack(mssim)))
|
|
np.save(f"{args.plot_data}_{args.iterations}_{i:08.3f}.npy", np.column_stack((dec[3], data)))
|
|
|
|
|
|
|
|
|