Bio7_Classification

Open an image in ImageJ and follow Button 1-4 for a classification workflow which trains and classifies images with R (scripts are available in the R directory) in a non-blocking job. For an unsupervised classification (e.g., k-means) just use the classification action (4). RGB images can be converted to a different color space (see option HSB Stack, LAB Stack) and selected channels from multichannel images or stacks can be extracted individually. Several features can be enabled in the default tab which will be added to the default image layers. A comma separated text argument adds filter images of different radius or applies special settings for edge algorithms like Difference of Gaussian, Lipschitz, Gabor, Convolve. For some edge detection methods a ';' separator can be set, too, for different sets of edge settings (Difference of Gaussian, Lipschitz, Gabor, Convolve) resulting in one image layer each. All settings for classification can be stored or reopened with the 'Load/Save Configuration' actions in a simple text file (simply drag the file on the GUI interface to load it!). In the Settings tab the path to the R (training and classification) and the ImageJ import macro scripts can be set if necessary (or easier simply change the default scripts). It is also possible to enable a directory dialog (see option 'Use Directory Dialog') for the classification (step 4) of images in a folder and it's subfolders (lists the image files recursively) instead of using selected images with the file dialog (for multiple files). In addition the data transfer type to R can be selected (whereat most filters require a double transfer - this action will set the transfer type of images, too). It is also possible to open the default hidden feature stack which will be transfered to R (option 'Open Feature Stack') and transfer signatures according to the group membership of the ROI (option 'Use Group Signature'). With the "Create Classification Project" action a classification Bio7 project can be created with a selectable folder structure. The default R templates can be copied to the new project location for custom scripts (paths have to be adjusted and stored in a configuration file!). If the 'Selection Preview (Train)' option is enabled a classification overlay will be generated on top of the source image with size and location of the selection. A retraining of the dataset for the preview can be enabled if the 'Retrain for Preview' option is selected. The ROI overlay will be updated dynamically if you change or move the ROI selection. Please disable the preview option (a ROI listener) or close the GUI if you recompile the plugin! The preferred LUT and opacity can be selected, too. The enabled option "Show in ImageJ" transfers the data back to ImageJ and can be disabled if the classification result should be stored with the R scripts. With the option "Apply ImageJ Macro" the classified image can be post-processed with an ImageJ macro, e.g., to extract and measure particles or identify image objects. Until now Multichannel images (e.g. RGB) and Grayscale images or stacks (8-bit, 16-bit, 32-bit) can be classified. It is also possible to import images with an ImageJ macro (e.g. Landsat 8 images, see ImageJ macro example!). For convenience images and stored ROI Manager files can be dropped on the ImageJ-Canvas view to open them. In addition a saved R workspace file can be opened, too, by dropping it on the main toolbar of Bio7. Finally a stored GUI configuration file can be opened by dropping it directly on the classification GUI.