Description

Performs pixel classification using the WEKA Trainable Segmentation plugin.

This module loads a previously-saved WEKA classifier model and applies it to the input image. It then returns the multi-channel probability map.

Image stacks are processed in 2D, one slice at a time.

Parameters

Input image: Image to apply pixel classification to.

Convert to RGB: Converts a composite image to RGB format. This should be set to match the image-type used for generation of the model.

Output image: Output probability map image.

Output bit depth: By default images will be saved as floating point 32-bit (probabilities in the range 0-1); however, they can be converted to 8-bit (probabilities in the range 0-255) or 16-bit (probabilities in the range 0-65535). This is useful for saving memory or if the output probability map will be passed to image threshold module.

Output single class: Allows a single class (image channel) to be output. This is another feature for reducing memory usage.

Output class: Class (image channel) to be output. Channel numbering starts at 1.

Path type: Method to use for generation of the classifier filename:

Generic format: Format for a generic filename. Plain text can be mixed with global variables or metadata values currently stored in the workspace. Global variables are specified using the "V{name}" notation, where "name" is the name of the variable to insert. Similarly, metadata values are specified with the "M{name}" notation.

Available metadata fields: List of the currently-available metadata values for this workspace. These can be used when compiling a generic filename.

Classifier file path: Path to the classifier file (.model extension). This file needs to be created manually using the WEKA Trainable Segmentation plugin included with Fiji.

Simultaneous slices: Number of image slices to process at any given time. This reduces the memory footprint of the module, but can slow down processing.

Tile factor: Number of tiles per dimension each image will be subdivided into for processing. For example, a tile factor of 2 will divide the image into a 2x2 grid of tiles. This reduces the memory footprint of the module.