Deconvolving Synthetic Data

Given here is an example of a workflow for using our deconvolution algorithm with synthetic data. The same workflow can be run at once by changing the relevant parameters in the script example_synthetic.py and running it as

python example_synthetic.py

Creation of Noisy Synthetic Data

First step is to create the synthetic data using the parameters detailed in parameters. For example:

from synthetic_data.synthetic_data import SyntheticData

sd_params = dict(size=200, frames=50, spotsize=3, numpoints=50, motion=3, brightness=0.1)
SD = SyntheticData(**sd_params)

testdata = SD.create_testdata()

To convolve with a PSF, continue as follows:

from psf.psf import PSF
import numpy as np
from util.timedependence import convolve_fft

psf_params = dict(type='widefield', lambdaEx=561, lambdaEm=609, numAper=1.4,
                  magObj=100, rindexObj=1.518, ccdSize=6540, dz=0, xysize=200, nslices=1,
                  depth=0, rindexSp=1.518, nor=0)
pf = PSF((200, 200), **psf_params)


psf_big = np.zeros(testdata.shape)
psf_big[testdata.shape[0] // 2, :, :] = pf.data
testdata_psf = convolve_fft(testdata, psf_big)
  

Finally, noise can be added to the synthetic data.

noisy_p = SD.add_poisson_noise(testdata_psf, p)
noisy_g = SD.add_gaussian_noise(noisy_p, g)
  

where p and g control the level of Poisson and Gaussian noise, respectively.

Deconvolution

For the deconvolution, the deconvolution parameters lambda, lambda_t, epsilon and maxit are needed. See usage. Of course it is necessary to use the same PSF (or generate one with the same parameters), as was used to degenerate the data earlier.

from util.deconvolution import Deconvolution

Dec = Deconvolution(pf.data, noisy_g, l=0.1, lt=0.1, eps=0.001, maxit=1)
result = Dec.deconvolve()                
  