CANDLE-J is a 3D image denoising software designed as an ImageJ (64-bit) plugin and an open source alternative to the CANDLE MATLAB program. CANDLE-J is very adept at processing deep in vivo multiphoton microscopy images where the signal to noise ratio (SNR) is low. The Collaborative Approach for eNhanced Denoising under Low-light Excitation (CANDLE) has been tested on synthetic data, images acquired on microspheres and in vivo images.
The denoising algorithms are written in Jython, Java and C. The C methods are accessed with the aide of the Java Native Access library. The Optimized Non Local Means Filter (ONLM) is the bottleneck of the CANDLE-J software and hence the ONLM algorithm is written as a multithreaded C program to improve performance. Additionally, the two C programs were compiled with the GCC optimizing option
-O3 turned on. The
-O3 option makes the compiler strive to improve the performance of the object code.
CANDLE-J is an ImageJ (64-bit) plugin and requires ImageJ to be installed on the computer. If ImageJ is not already installed, click here to download the appropriate version. In the top header of this repository, click release. Depending on your operating system, either download the
Linux.tar.gz file from the listing. Unzip the downloaded file, and place the resultant
CANDLE-J folder in the
plugins folder of your local ImageJ directory. Open ImageJ (restart ImageJ if it is already open) and CANDLE-J should be available to use from the
Plugins dropdown menu.
For a Mac OS X; CANDLE-J should be used with the default
ImageJ(64-bit) instead of the version named
ImageJ32(32-bit). Both come bundled with the downloaded ImageJ zip file.
In a Linux machine, use the version of ImageJ that appears with the default Linux executable icon: . Using the wrong version causes an additional statement to appear in the ImageJ log window,
Error: Could not load pre-computed correspondence table.
1- Open ImageJ
2- Go to the Plugins dropdown menu.
3- Hover your mouse on CANDLE-J and click CANDLE.
5 - A GUI to select filtering parameters will appear. Also a dialog box will show up to keep the user updated regarding the progress of the program.
- First, select the smoothing parameter (beta). This parameter controls the amount of denoising to be applied. Generally, values between 0.1 and 0.4 are fine for visualization. Higher values may be useful for segmentation or registration purposes.
- Second, select the patch radius. A radius of 1 voxel (i.e., patch of 3x3x3 voxels) is usually appropriate for a greater view of fine structures. A radius of 2 voxels (i.e., patch of 5x5x5 voxels) should be favoured for a restricted field of view of large structures.
- The radius of the search volume can also be changed. However, it is not recommended to increase it (higher computational time).
6- Once CANDLE-J has run its course, the ouput image will be displayed (can be manipulated further with ImageJ) along with a save dialog box.
Note - A noisy sample image stack is provided to test architecture compatibility and ensure that CANDLE-J is working properly. To download the noisy image stack, go to the
Sample Images folder located in the list of files above and download
NoisyImage.tif. You can also download
DenoisedImage.tif from the same folder to see how the noisy image stack is supposed to look like after being denoised (using the default parameters) by CANDLE-J.
* * *
Haider Riaz Khan
Montreal Neurological Institute
3801, Rue University, MP121 (lab)
Montreal, QC H3A 2B4