Interactive Learning and Segmentation Tool Kit
Christoph Sommer, Christoph Straehle, Ullrich Köthe and Fred A. Hamprecht
ilastik has a convenient mouse interface for labeling an arbitrary number of classes in the images. These labels, along with a set of generic (nonlinear) image features, are then used to train a Random Forest classifier. In the interactive training mode, ilastik provides real-time feedback of the current classifier predictions and thus allows for targeted training and overall reduced labeling time. In addition, an uncertainty measure can guide the user to ambiguous regions of the data. Once the classifier has been trained on a representative subset of the data, it can be exported and used to automatically process a very large number of images.
The features are computed in the full 2D/3D/4D pixel neighborhoods, depending on the available data. While the provided set of features includes popular color, edge and texture descriptors, the plug-in functionality allows advanced users to add their own problem-specific features.
Feature computation and classifier prediction are multi-threaded and fully exploit modern multi-core machines.
ilastik is an open source project and is released under the BSD license.
In summary, the main features of ilastik are:
- Support of up to 4D (3D multi-spectral) data
- Learning of multiple classes
- Generic features cope with a wide array of local image structures and textures
- Features can be computed in all dimensions
- Batch mode allows automated processing of new images of the same kind
- Multi-threaded execution reduces processing time
- No programming expertise needed to interactively explore images and train a classifier
- Plug-in mechanism offers extensibility
- Friendly open source developer community :-)
Last update: 26.10.2010, 20:31