imodfindbeads(1) imodfindbeads(1)
NAME
imodfindbeads - Find gold particles in images
SYNOPSIS
imodfindbeads options in_image out_model
DESCRIPTION
Imodfindbeads finds gold particles or other circular densities (beads)
in images by a combination of cross-correlation and other methods. It
starts by correlating with a model of a spherical bead of a specified
size, then forms an average out of the most strongly-correlating subset
and repeats the procedure by correlating with the average. It analyzes
the distribution of correlation strengths to find the strength that
best separates the particles of interest from similar densities. The
positions of the beads are stored in an IMOD model along with strengths
for each. The points can then be visualized in 3dmod, and with the
help of the Bead Fixer module, the threshold can be adjusted and points
below threshold can be deleted.
Rather than cross-correlating with a model or averaged bead, the pro-
gram applies an edge-detecting filter (Sobel, by default) to both the
images and the reference, and correlates the filtered images. This
method improves the detectability of the beads and may improve the
accuracy of the center positions. However, it only works well for
beads in a certain size range, so the program first scales the images
to bring the beads to a specified size (8, by default). The peaks in
this correlation are the set of candidate positions for the beads.
At each position, the program then computes an integral of the bead
density relative to the background in an annulus around the bead. The
program can then work with three measures of peak strength. One is the
strength of the Sobel-filter correlation (which includes a component
based on the density of the bead, a factor lost when using a normalized
correlation coefficient). The second is the integrated density, and
the third is the geometric mean of the first two. Whichever measure is
chosen, it is scaled so that the maximum value is 1.
Correlation with a simple item like a bead always produces many more
peaks than actual beads, but a histogram of peak strength generally
shows a dip between actual beads and spurious peaks. The program thus
computes a histogram and smooths it with kernel smoothing, whereby a
narrow distribution function instead of a single point is added into
the histogram at every peak position. The width of this function is
the kernel width, referred to as H in program output. The program
tries a series of widths, from 0.2 downward, until it finds a dip in
the histogram; then it computes it again with a kernel width of 0.05 in
order to locate the dip more accurately.
After the initial correlation with a model bead, the program uses the
histogram analysis to select beads to average as a template for the
second round correlation. If the analysis fails, it is possible to
bypass it by entering a relative peak strength to use as a criterion
for selecting beads. After the second round of correlations, the loca-
tion of the dip is used to determine which points to output in the
model. The default is to output a number of points below the dip so
that the user can check and adjust the threshold if necessary. How-
ever, with the -store option, you can output just the points above the
dip or a certain fraction of the strongest peaks above the dip. Or, if
the histogram analysis fails, this option can be used to bypass it and
specify the actual peak strength to use as the criterion for output.
The most significant options described below are: -size to specify the
bead size, -area to specify a model with contours around areas to ana-
lyze, -light if beads are light on a dark background, -store to control
storage of peaks in the model as just described, -spacing to allow
points closer together than the bead size, -sections to select the sec-
tions to be analyzed, and -maxsec to set how many of them will be ana-
lyzed in one group. In addition, if images are noisy, it may be help-
ful to use linear interpolation with -linear and add some filtering
with either -kernel or -rad2 and -sig2. There are many other options
that were added during program development and can be ignored.
Options
Imodfindbeads uses the PIP package for input (see the manual page for
pip). Options can be specified either as command line arguments
(with the -) or one per line in a command file (without the -):
-input OR -InputImageFile File name
Name of input image file. If it is not entered with this option
it must be entered with the first non-option argument.
-output OR -OutputModelFile File name
Name of output model file. If it is not entered with this
option it must be entered with the second non-option argument.
-filtered OR -FilteredImageFile File name
Output file for images after they have been scaled and Sobel
filtered. The coordinate system in the header will be congruent
with that of the original image file so that the model can be
displayed on it.
-area OR -AreaModel File name
Model with contours enclosing areas to analyze. More than one
contour can be included on each section. If there are no con-
tours on a section within the range being analyzed, the entire
section will be used.
-add OR -AddToModel File name
Model to append to. After the analysis of histograms and
thresholds, detected points will be eliminated if they are
within the criterion spacing of a point in an existing object
(where closed contour objects are excluded). Points will be
stored in new objects.
-ref OR -ReferenceModel File name
Model with all beads marked in defined areas, used for determin-
ing the accuracy of the bead detection.
-boundary OR -BoundaryObject Integer
Object in reference model with contours around the areas where
beads have been fully marked.
-size OR -BeadSize Floating point
Size of beads in pixels, a required entry. A model bead of this
size is constructed for the first round of correlation, and this
size together with the scaled size determine how much images
will be scaled for filtering.
-light OR -LightBeads
Beads are light on dark background
-scaled OR -ScaledSize Floating point
Size of beads in images scaled down for filtering; this entry
together with the bead size determine the amount of scaling. In
tests, values of 7 to 10 have been found to give the best detec-
tion with Sobel filter. The default value is 8.
-interpmin OR -MinInterpolationFactor Floating point
When images are being scaled down, this entry determines how
much of the scaling down must be done with interpolation instead
of binning. Interpolation will preserve high frequency informa-
tion better than binning does but may amplify noise. The
default value is 1.4, which means that data will not be binned
unless they are being scaled down by a factor of at least 2.8.
This entry makes no difference unless images are being scaled
down by more than a factor of 2.
-linear OR -LinearInterpolation
Use linear instead of cubic interpolation; this option will help
to reduce noise.
-center OR -CenterWeight Floating point
The weighting for the center pixel in the edge detecting filter;
1 and 2 correspond to Prewitt and Sobel filters, respectively;
the default is 2.
-box OR -BoxSizeScaled Integer
Box size for correlating and averaging beads in scaled down
images. The default is 3 times the scaled size plus 2.
-threshold OR -ThresholdForAveraging Floating point
Threshold relative peak strength or number of beads for averag-
ing. With a non-zero entry, selected beads from the first round
of filtering and correlation are averaged to produce a reference
for a second round. If a negative value is entered, the program
will analyze the histogram of peak strengths and find the dip
indicating the best boundary between actual and false beads.
The value has 4 different meanings depending on the range:
Greater than 1: an absolute number of beads with the strongest
peaks
Between 0 and 1: minimum relative peak strength
Between 0 and -1: negative of strongest fraction of peaks
above histogram dip
-2: 1/4 of way from histogram dip to histogram peak (the
default)
-store OR -StorageThreshold Floating point
Threshold relative peak strength for storing peaks in model.
With a value of 0 (the default), the program will find the dip
in the histogram of peak strengths, find the mean and SD of the
strengths above the dip, and store all of the beads above the
dip plus additional ones below the dip. The latter will be up
to the 5 SD's below the mean or up to the number of ones above
the dip. Enter a number between 0 and 1 to specify a relative
strength above which peaks will be stored. Enter a negative
number to specify the number to store as a fraction of the num-
ber above the histogram dip (e.g., -1 for all points above the
dip, -0.33 for the strongest 1/3 above the dip, -1.33 for all
above plus 1/3 that many below the dip.)
-bkgd OR -BackgroundGroups Floating point
After finding peaks, the program will sort the peaks based on
the background density into the number of groups given by this
entry, as long as there are at least 100 peaks in each group.
The histogram will be analyzed separately for each group to find
the dip, and then the peak strengths will be scaled so as to
superimpose the dip values. This scaling should make a single
threshold value work better across a range of intensities. The
default value is 4; enter 0 to prevent this analysis.
-annulus OR -AnnulusPercentile Floating point
By default, the program will use the mean in an annulus around
the bead as the background for the analysis by groups. This
entry specifies a percentile of the pixel values to use instead
(e.g., 0.5 for the median).
-peakmin OR -MinRelativeStrength Floating point
Minimum relative peak strength, after any background scaling,
for keeping a peak in the analysis. Too many weak peaks can
prevent a dip from showing up in the smoothed histogram of
strengths. The default is 0.1, or 0.05 if a threshold for aver-
aging is being found from histograms.
-spacing OR -MinSpacing Floating point
Minimum spacing between peaks as a fraction of the bead size.
When two peaks are closer than this distance apart, the weaker
one is eliminated. The default is 1, but values of 0.8 to 0.9
are helpful for getting a more complete set of beads.
-sections OR -SectionsToDo List of integer ranges
List of sections to run. Comma-separated ranges can be entered;
sections are numbered from 0. By default, all sections will be
analyzed.
-maxsec OR -MaxSectionsPerAnalysis Integer
Maximum number of sections to include in one analysis. With
this entry, the list of sections will be divided into groups no
bigger than this size.
Each group will be analyzed separately and results will be
stored in a separate model object with its own threshold value.
By default, all sections are analyzed together.
-remake OR -RemakeModelBead
Start with a model bead for each separate analysis when more
than one group of sections is being analyzed. The default is to
use the average from the previous group for the first round of
correlation on a group.
-guess OR -MinGuessNumBeads Integer
A guess for the minimum number of beads per section. This entry
may sometimes be required to help the program find a dip in the
histogram, especially if there are very few beads.
-measure OR -MeasureToUse Integer
Measure to use for peak strengths: 0 for the correlation peak, 1
for the integral of density above the background, 2 for the geo-
metric mean of these two. The default is 1; integrals were
slightly better in test data sets.
-kernel OR -KernelSigma Floating point
Sigma for real-space smoothing with a Gaussian kernel (in pix-
els). The smoothing is with a 3x3, 5x5 or 7x7 kernel whose
coefficients are proportional to a Gaussian with the given sigma
centered on the central pixel. This smoothing is applied before
the image is scaled for filtering. The default is 0.85, which
is equivalent to the simple smoothing filter in Clip and
3dmod.
-rad1 OR -FilterRadius1 Floating point
Low spatial frequencies in the cross-correlation will be attenu-
ated by a Gaussian curve that is 1 at this cutoff radius and
falls off below this radius with a standard deviation specified
by FilterSigma2. Spatial frequency units range from 0 to 0.5.
Use FilterSigma1 instead of this entry for more predictable
attenuation of low frequencies.
-rad2 OR -FilterRadius2 Floating point
High spatial frequencies in the cross-correlation will be atten-
uated by a Gaussian curve that is 1 at this cutoff radius and
falls off above this radius with a standard deviation specified
by FilterSigma2.
-sig1 OR -FilterSigma1 Floating point
Sigma value to filter low frequencies in the correlations with a
curve that is an inverted Gaussian. This filter is 0 at 0 fre-
quency and decays up to 1 with the given sigma value. However,
if a negative value of radius1 is entered, this filter will be
zero from 0 to |radius1| then decay up to 1.
-sig2 OR -FilterSigma2 Floating point
Sigma value for the Gaussian rolloff below and above the cutoff
frequencies specified by FilterRadius1 and FilterRadius2
-verbose OR -VerboseKeys Text string
Key letters for verbose output (1 for general, p for peak, f for
first, l for last, e for every hist)
-param OR -ParameterFile Parameter file
Read parameter entries as keyword-value pairs from a parameter
file.
-StandardInput
Read parameter entries from standard input
AUTHOR
David Mastronarde
SEE ALSO
beadtrack
BUGS
Email bug reports to mast at colorado dot edu.
BL3DEMC 4.5.6 imodfindbeads(1)