Home
Welcome to the optimisme wiki!
OPTIMIsation Stochastique en imagerie MultispectralE,
The scientific project was at first a french project for young researchers financed by the Gdr ISIS in 2013. It then continued and extended with the financial support of CNRS within the project Imag’In in 2015-2016 under the name OPTIMISME for OPTIMIsation Stochastique en imagerie MultispectralE i.e. Stochatic Optimization in Multispectral Imagery. This project focuses on the restoration of two-photon microscopy data. It is composed of two parts: a theoretical approach and its application to two-photon microscopy data. In the theoretical part, proximal algorithms have been developed, one of them is based on a Majorization-Minimization procedure. In the applicative part, the developed algo- rithms were used to restore two-photon microscopy images. Furthermore this approach included an analysis of the noise and the non-stationarity of the PSF.
Modern approaches of inverse problems resolution in the field of multi- or hyper-spectral imaging are based on variational formulations. The goal of this study is to provide a generalization of parallel methods using recent advances of stochastic optimization, for the treatment of massive big data. The algorithms provided solve the multispectral deconvolution problem in two-photon microscopy.
The in-vivo observation of a mouse brain by a two-photon microscope leads to direct images at cellular scales, which are degraded by blur and noise and need restoration. This problem yields an inverse problem that was targeted by the OPTIMISME project. The issue of obtaining the PSF of the two-photon microscope was solved by observing fluorescent micro beads (of diameter 0.5 μm) at different depths. The PSF varies with the depth (z coordinate). The OPTIMISME-ImageJ plugin implements the Majorization- Minimization proximal algorithm described in [1] to restore an image that has been blurred by a known PSF.
ImageJ is an image processing software dedicated to multidimensional scientific images. ImageJ can be easily extended adding plugins and scripts. ImageJ web site:
ImageJ gives the possibility to user to develop and install one’s own plugin. The following page explain how to install an ImageJ plugin :
Libraries used for the OPTIMISME algorithm are:
- edu.emory.mathcs.jtransforms For 3D and 2D FFT and IFFT. On license BSD clause 2. Non contaminant license.
- org.ejml a part of ejml library for matrix calculus and pseudoInvers function. On Apache license Non contaminant license.
- org.apache.commons.math3 a part of apache common library for tricubic interpo- lation. on Apache license Non contaminant license.
The OPTIMISME plugin requires version 1.48 or higher of ImageJ (or Fiji) and java 8 or higher. (To check your installation click on Help>About ImageJ.., the java version must be 1.8.0 or higher).
The plugin OPTIMISME is available under the following formats.
To install it,
- Add the directory OPTIMISME in the plugin directory of ImageJ (or Fiji) (with all the *class files). Then start ImageJ (or Fiji).
- If the plugin item "OPTIMISME" does not appear in the menu : Go to Help>Refresh Menus or (better) restart ImageJ (or Fiji).
If you are using Fiji, it is possible to install the plugin through the public distribution using the install plugin menu (Plugins>Install PlugIn..).
For earlier versions of ImageJ or Java, it may be necessary to compile the plugin. The java files are available in the OPTIMISME folder. To compile them,
- Add the directory OPTIMISME in the plugin directory of ImageJ.
- Rename the file optimisme.txt as optimisme_.java.
- Start ImageJ.
- Click on Plugins > Compile and Run.
- Select the file optimisme_.java.
- If the plugin item "OPTIMISME" does not appear in the menu : Go to Help>Refresh Menus or (better) restart ImageJ.
The Plugin is accessible through the “Plugins” menu of ImageJ. The OPTIMISME plugin
assumes that at least two images are opened: the image to restored and the PSF. Note
that the present version is established with one single PSF.
Launch the OPTIMISME Plugin as follows:
In the first window of the plugin OPTIMISME, the user specifies the input image, the channel that needs restoration, and the PSF image. Others parameters of the algorithm described hereafter, may also be specified (see [1] for more details).
In the window of OPTIMISME Plugin, the Input parameters have default values, that can be modified by the user.
- NbIt: number of iteration of the algorithm that are performed.
- regul: regularization coefficient for XY coordinate (space)
- T: smoothing parameter for the XY regularization term
- regulZ: regularization coefficient for the Z coordinate (depth)
- phixy: choice of the XY regularization function
- phiz: choice of the Z regularization function
- TZ: smoothing parameter for the Z regularization term
- eta: regularization parameter to force the solution to be positive
- Channel number: in case the input image contains several channels, the user can select the channel to restore.
Before you launch OPTIMISME, you have to open two images corresponding to 1) the image to be processed and 2) the PSF to be used. OPTIMISME needs as input file two images (image and PSF). ImageJ accepts different formats (tiff format, lsm directly issued from microscope...).
- File>Open open at least two files one for Image and one for the PSF. OPTIMISME assumes 3D files, images with a depth dimension.
The algorithm first compares the resolutions of the image and the PSF: if resolutions are different the PSF is re-sampled to the resolution of the image; at the end of this first stage the re-sampled PSF is plotted (it may then be stored by the user for a later use). The plugin then proceeds with the actual restoration algorithm (see [1] for more details), and finishes by plotting of the processed image.
The OPTIMISME plugin outputs at least one image
- At the end of the computations an image is opened: it is the deconvolved image, the user can save it using the File>Save as menu.
- When the input images (image and PSF) do not have the same resolution, the algorithm first re-samples the PSF; then, at the end of this first stage, the program displays the re-sampled PSF with the same resolution as the input image. The user can save it for further runs of the program and use it for all the images having the same resolution using the File>Save as menu.
The java class of algorithm Majorization/Minimization in ImageJ plugin, and other dependencies
. /ImageJ/plugins/OPTIMISME/
│-- org.jar
│-- fftj.jar
└─── doc/
│-- index.html
│-- ...
│-- optimisme_.java
└── data
│-- image_synth.tiff
│-- psf_synth.tiff
│-- psf_synth_subsampled.tiff
└── optimisme
│-- MM.java
│-- PSFPreparator.java
│-- MMCal.java
└── test
│-- AllTests.java
│-- MMCalTest.java
│-- MMTest.m
│-- PSFPreparatorTest.m
Tests files are in the test folder. The coverage by Emma leads to :
| source files | file size | %CPU | %MEM | %TIME |
|---|---|---|---|---|
| stark1_840*2 | 342 (512512261) | 100 | 61 | 763 -stop |
| stark1_450*2 | 225 (512512172) | 100 | 61 | 763 -stop |
| stark1_840* psf_synth | 167 (512512128) | 100 | 7.4 | 22:20 |
| stark1_840* psf_synth | 167 (512512128 | 324 | 81 | 3:00 |
[1] E. Chouzenoux, L. Lamasse, S. Anthoine, C. Chaux, A. Jaouen, I. Vanzetta, and F. Debarbieux. Approche variationnelle pour la déconvolution rapide de données 3D en microscopie biphotonique. In Actes du 25e colloque GRETSI, Lyon, France, September 2015.
- Caroline CHAUX
- Sandrine ANTHOINE
- Dominique BENIELLI
