def doFit(fit_function, fit_data, max_iterations):
save_residual = False
if save_residual:
resid_dax = daxwriter.DaxWriter("residual.dax",
fit_data.image.shape[0],
fit_data.image.shape[1])
for i in range(max_iterations):
if save_residual:
resid_dax.addFrame(fit_data.residual)
done = findPeaks(fit_data)
fit_function(fit_data)
updateBackgroundCutoff(fit_data)
if not(done):
break
#cutoff = background + threshold * std
if save_residual:
resid_dax.close()
if (type(fit_data.peaks) == type(numpy.array([]))):
fit_data.peaks[:,util_c.getXCenterIndex()] -= fit_data.pad_size
fit_data.peaks[:,util_c.getYCenterIndex()] -= fit_data.pad_size
# TODO: a final phase of refitting after removal of bad peaks?
return [fit_data.peaks, fit_data.residual]
def analyzeImage(self, new_image, save_residual = False, verbose = False):
pad_image = fitting.padArray(new_image, self.peak_finder.margin)
self.peak_finder.newImage(pad_image)
self.peak_fitter.newImage(pad_image)
#
# This is a lot simpler than 3D-DAOSTORM as we only do one pass,
# hopefully the compressed sensing (FISTA) deconvolution finds all the
# peaks and then we do a single pass of fitting.
#
if True:
peaks = self.peak_finder.findPeaks()
[fit_peaks, residual] = self.peak_fitter.fitPeaks(peaks)
#
# This is for testing if just using FISTA followed by the center
# of mass calculation is basically as good as also doing the
# additional MLE spline fitting step.
#
# The short answer is that it appears that it is not. It about
# 1.3x worse in XY and about 4x worse in Z.
#
else:
fit_peaks = self.peak_finder.findPeaks()
# Adjust z scale.
z_index = utilC.getZCenterIndex()
z_size = (self.peak_fitter.spline.shape[2] - 1.0)
status_index = utilC.getStatusIndex()
fit_peaks[:,z_index] = z_size*fit_peaks[:,z_index]
# Mark as converged.
fit_peaks[:,status_index] = 1.0
residual = None
#
# Subtract margin so that peaks are in the right
# place with respect to the original image.
#
fit_peaks[:,utilC.getXCenterIndex()] -= float(self.peak_finder.margin)
fit_peaks[:,utilC.getYCenterIndex()] -= float(self.peak_finder.margin)
return [fit_peaks, residual]
def getPeaks(self, threshold, margin):
fx = self.getXVector()
# Get area, position, height.
fd_peaks = fdUtil.getPeaks(fx, threshold, margin)
num_peaks = fd_peaks.shape[0]
peaks = numpy.zeros((num_peaks, utilC.getNResultsPar()))
peaks[:, utilC.getXWidthIndex()] = numpy.ones(num_peaks)
peaks[:, utilC.getYWidthIndex()] = numpy.ones(num_peaks)
peaks[:, utilC.getXCenterIndex()] = fd_peaks[:, 2]
peaks[:, utilC.getYCenterIndex()] = fd_peaks[:, 1]
# Calculate height.
#
# FIXME: Typically the starting value for the peak height will be
# under-estimated unless a large enough number of FISTA
# iterations is performed to completely de-convolve the image.
#
h_index = utilC.getHeightIndex()
#peaks[:,h_index] = fd_peaks[:,0]
for i in range(num_peaks):
peaks[i, h_index] = fd_peaks[i, 0] * self.psf_heights[int(
round(fd_peaks[i, 3]))]
# Calculate z (0.0 - 1.0).
peaks[:,
utilC.getZCenterIndex()] = fd_peaks[:, 3] / (float(fx.shape[2]) -
1.0)
# Background term calculation.
bg_index = utilC.getBackgroundIndex()
for i in range(num_peaks):
ix = int(round(fd_peaks[i, 1]))
iy = int(round(fd_peaks[i, 2]))
peaks[i, bg_index] = self.background[ix, iy]
return peaks
average_psf = numpy.zeros((4*aoi_size,4*aoi_size))
curf = 1
peaks_used = 0
total = 0.0
[dax_x, dax_y, dax_l] = dax_data.filmSize()
while (curf < dax_l) and (peaks_used < min_peaks):
# Select localizations in current frame & not near the edges.
mask = (i3_data['fr'] == curf) & (i3_data['x'] > aoi_size) & (i3_data['x'] < (dax_y - aoi_size - 1)) & (i3_data['y'] > aoi_size) & (i3_data['y'] < (dax_x - aoi_size - 1))
xr = i3_data['x'][mask]
yr = i3_data['y'][mask]
ht = i3_data['h'][mask]
# Remove localizations that are too close to each other.
in_peaks = numpy.zeros((xr.size,util_c.getNResultsPar()))
in_peaks[:,util_c.getXCenterIndex()] = xr
in_peaks[:,util_c.getYCenterIndex()] = yr
in_peaks[:,util_c.getHeightIndex()] = ht
out_peaks = util_c.removeNeighbors(in_peaks, aoi_size)
print curf, in_peaks.shape, out_peaks.shape
# Use remaining localizations to calculate spline.
image = dax_data.loadAFrame(curf-1).astype(numpy.float64)
xr = out_peaks[:,util_c.getXCenterIndex()]
yr = out_peaks[:,util_c.getYCenterIndex()]
ht = out_peaks[:,util_c.getHeightIndex()]
for i in range(xr.size):
peaks_used = 0
total = 0.0
[dax_x, dax_y, dax_l] = dax_data.filmSize()
while (curf < dax_l) and (peaks_used < min_peaks):
# Select localizations in current frame & not near the edges.
mask = (i3_data['fr'] == curf) & (i3_data['x'] > aoi_size) & (
i3_data['x'] < (dax_x - aoi_size - 1)) & (i3_data['y'] > aoi_size) & (
i3_data['y'] < (dax_y - aoi_size - 1))
xr = i3_data['x'][mask]
yr = i3_data['y'][mask]
ht = i3_data['h'][mask]
# Remove localizations that are too close to each other.
in_peaks = numpy.zeros((xr.size, util_c.getNResultsPar()))
in_peaks[:, util_c.getXCenterIndex()] = xr
in_peaks[:, util_c.getYCenterIndex()] = yr
in_peaks[:, util_c.getHeightIndex()] = ht
out_peaks = util_c.removeNeighbors(in_peaks, aoi_size)
print curf, in_peaks.shape, out_peaks.shape
# Use remaining localizations to calculate spline.
image = dax_data.loadAFrame(curf - 1).astype(numpy.float64)
xr = out_peaks[:, util_c.getXCenterIndex()]
yr = out_peaks[:, util_c.getYCenterIndex()]
ht = out_peaks[:, util_c.getHeightIndex()]
for i in range(xr.size):