readme.txt

Edge detection in microscopy images using curvelets

directions are shown in the "Directions to show" list box, but level 4
has 32 directions, then selecting direction 1 in the list box will
include directions 1 and 2 on level 4.

Symmetric directions: Checking this box, forces the selection of
directions to be symmetric, i.e. when one direction is selected, the
direction opposite to it (shifted by 180 degrees) is also included. It
is often wise to enable this option, because it ensures that
reconstructed images will be real.

Domain: This drop-down box determines what will be shown in the image
axes next to it. The choices are "Original image", "Reconstructed
image", "Error" (reconstructed minus original) and "Coefficients" (the
value of the curvelet coefficients on the coarsest selected level and
first selected direction for all space points - this is not equal to
the reconstruction). Original, reconstructed and error images can be
viewed either in physical space or in Fourier space. And the
coefficients can be viewed either for the original image or for the
reconstructed (and thresholded) image.

Values: This determines which values are shown in the image axes next
to it - real part, imaginary part, absolute value or logarithm of
absolute values.

Thresholding: In this box, the thresholding of curvelet coefficients
for reconstruction can be specified. Selecting "Threshold" in the
drop-down keeps only coefficients with absolute value larger than the
value specified in the edit box. Selecting "Number" keeps the largest
coefficients, but only the number of coefficients specified in the
edit box. The "Percentage" option keeps the largest fraction specified
in the edit box, i.e. writing 0.1 keeps the 10% largest coefficients.

Show in separate window: Clicking these buttons opens a new figure
window and displays the image currently displayed, but in a new window
- useful for comparisons, and for printing and saving images.

Export data: Exports current internal program data to a variable
"cviewdata" in the matlab workspace.

l^2 error and PSNR: Show the error of the reconstruction as l^2-error,
||im - rec||_2, and as
 PSNR (= 20 * log10((max(im) - min(im)) / ||im - rec||_2) 


4.1.2.2 The Toolbar

The toolbar contains the usual MATLAB Print, Zoom and Pan tools, but
also the following tools:

Link axes (the up-down arrow): Links the zoom and pan operations in
the two axes (useful for comparing original and reconstructed images)

Plot directions (the red star-like shape): Turns the bottom axes into
a polar plot, when the user clicks in the top image (displaying the
original or reconstructed image). The polar plot shows the magnitude
of curvelet coefficients in different directions for the selected
detail levels at the position clicked in the image. By the check-boxes
that appear to the left, the user can turn the polar plot into a
cartesian plot, specify a sliding average over a specified number of
neighboring curvelets before plotting, and choose to sum coefficients
on all the selected levels.

Norm of directions (the multi-colored disk): Displays the l^2-norm of
the curvelet coefficients at all levels and directions, computed
separately at each level and direction.


4.1.2.3 Menu items

The Tools menu contains two menu items: "Make movie" and "Batch
analysis...". The "Make movie" item opens the CMovieMaker window that
enables the user to create movies of a series of images, where
curvelet and/or wavelet transforms and thresholding can be applied to
each image (with similar options like in the CViewer GUI).
The "Batch analysis" item opens the "Batch window", where the user can
select a directory, and all images in this directory are loaded and
the current level, direction and thresholding options selected in the
CViewer window are applied to each image. For each image, a mat-file
is created, containing the same data as when clicking "Export data",
and the reconstructed image is saved in a subdirectory named Output,
and with a filename appening "_analyzed" to the original filename.



4.2 The EdgeDetect GUI

4.2.1 Basic features

The edgedetect GUI implements the algorithm presented in the paper
"T. Geback, P. Koumoutsakos; Edge detection in microscopy images using
curvelets; 2008"
for edge detection using the fast discrete curvelet transform.

The algorithm extracts a directional field with direction and
magnitude at each point given by the largest curvelet coefficient on
the chosen level. The resolution is given by the finest chosen level of curvelet
coefficients. The Canny edge detector is applied to trace along ridges
of the field magnitude to mark the edges.

4.2.2 All buttons and boxes in EdgeDetect GUI

4.2.2.1 Buttons and drop-downs

The "Directional field extraction" group contains options for
selecting which curvelet levels to use, and how to include them.
The "Levels" list box lets the user select which detail levels to
use. Checking "All curvelets" uses curvelets on the finest level
instead of wavelets. "Circular average size" specifies the size of the
averaging over directions before extracting curvelet magnitudes (1
means going 1 step left and 1 right, meaning a moving average of size
3). "Curvelet extent" specifies in the same way how many neighboring
positions should be averaged over (1 here means 1 left, 1 right, 1 up,
1 down - total 5). "# directional fields" means how many directional
fields should be extracted and used for the edge detection, the first
one being the largest coefficients, the second field the second
largest coefficients, if separated from the first by the number of
directions given in "Direction spacing".
Finally, "Filter type" specifies if we use the magnitude of the
curvelet coefficient, or the real part (corresponding to an even
filter), imaginary part (an odd filter), or sum of the real and
imaginary parts.

The "Edge detection" group lets the user specify thresholds for the edge
detection. More information about the Canny thresholds may be found by
writing "help edge" in the Matlab window. "Apply tophat transform"
applies a tophat-transform (original - opening of image) using a disk
of the specified radius as the structural element.
The "Sheet threshold" is used for extracting "sheets", which are areas
where the total curvelet magnitude (summed over all directions) is
large. Finally, "Exclude sheets from edges" prevents the marking of
edges in any areas that are marked as sheets.

The "Postprocessing" group provides some opportunities for
manipulating the extracted edges. An "Extension length" larger than 0
extends the found edges by at most the specified number of steps,
moving in the direction of the field. If another edge is encountered,
the travelled path between is included as an edge, connecting the two
edges. The global extension threshold gives the fraction of the
maximum magnitude that a pixel must exceed to be considered for
inclusion, while the local threshold is the fraction of the current
pixel value (that we are moving from).
"Thin edges" applies the matlab command "bwmorph('thin')" to the
edges. "Dilate and thin" applies the "bwmorph('dilate')" and the
"bwmorph('thin')" commands. "Remove spurs" applies "bwmorph('spur')".
"Skeletonize" uses bwmorph('thin', Inf) on the edges mapped to the
original image before displaying them.

Details about the edge detection and post-processing can be found in
curvecanny_multi.m.

In the "Display" group, the user can choose to display the image or
the curvelet magnitude image as background to the edges, choose to
display edges and/or sheets and/or the directional field (with small
arrows indicating direction and magnitude). Finally, the user can
select which of the extracted fields to show (if more than one).

By clicking "Export data", the user exports the current data
(including field, edges, sheets, etc) to the Matlab command window in
the variable edgedata.


4.2.2.2 The toolbar

The toolbar contains the standard Matlab tools for zooming and
panning.


4.2.2.3 The menu

The "Open file..." item opens a dialog box where the user can select
an image file to open. All image formats supported by Matlab's imread
function are supported, plus the .dcm DICOM format.

The "Batch analysis..." item opens the batch window, where the user
can select a directory from which all image files are read and
analysed using the current settings in the main window. For each
image, a Matlab mat-file is written in a subdirectory name Output,
containing similar data as exported by the "Export data" button. The
analyzed image is also written, corresponding to the options for
display specified in the main window. Some statistics is also
displayed in the Matlab command window for each analyzed file.



4.3 Other files and functions

For information on the other files and functions included in
CurveletUtils, see each of the separate files. The utils/ subdirectory
contains some simpler utilities that makes working with the Curvelet
data structure a bit simples. The src/ subdirectory contains source
for the mex-files used by the GUIs.




5 Citation information

If you use the EdgeDetect GUI or the underlying algorithm for any
sientific publication, please cite the following paper:
T. Geback, P. Koumoutsakos; "Edge detection in microscopy images using
curvelets", 2008