Feature List

This is a rough list of the major capabilities the Gerbil framework encompasses.

Input/Output

  • Support of a wide range of image formats, including ENVI, FITS, ESRI, TIFF, JPEG2k
  • Custom file format for easy import of data from Matlab, etc.
  • Processing of image data that does not fit into system memory
  • Reading and writing of label files, writing of spectral plots

Data Processing

  • Data reduction

    1. Region-of-interest selection, spatial masking
    2. Band range selection
    3. Interpolation, binning in the spectral domain

  • Normalization

    1. Data range calculation and histogramming
    2. Automatic, manual data normalization and range clamping
    3. Illuminant exchange

  • Preprocessing

    1. Flipping/transposition/rotation
    2. Apply logarithm, L2 normalization
    3. gaussian kernel denoising

  • Feature extraction

    1. Spectral gradient computation
    2. Pricipal Component Analysis (PCA)
    3. Spectral Angle Mapper, Spectral Information Divergence, SIDSAM computation

  • Label manipulation tools

    1. Find labels via supervised segmentation
    2. Create and manipulate labels via thresholding
    3. Edit labels with various label brushes
    4. Merge, delete etc. labels in a separate view

Visualization

  • Spectral Plots

    • Interactive Parallel-coordinates based spectral plots
    • Display distribution in feature spaces including:
      1. Raw image data
      2. L2-normalized spectra
      3. Spectral gradient
      4. Principal Component Analysis
    • Color-coding based on labels or true color/false color
    • Configurable drawing quality

  • Spatial Plots

    1. Display of image bands, spectral gradient bands with label overlay
    2. True-color display via CIE XYZ
    3. False-color display via PCA, Self-organizing Map (SOM)

  • Zoom/pan in all views, export views to file

Algorithms

  • Dimensionality reduction via SOM
  • Edge detection via RCMG, SOM
  • Segmentation
    1. Supervised (seeded) segmentation via Graph Cuts, Power Watersheds
    2. Superpixel segmentation
  • Material clustering via fast-adaptive mean shift (FAMS)
    • Accelerated clustering via superpixel-based FAMS
    • Accelerated clustering via SOM pre-processing

Interface

  • Powerful GUI interface using hardware-accelerated drawing on all major desktop platforms
  • Command-line interface for batch processing and automated evaluation
  • Internal C++-API for image data access and incorporation of new methods
  • Modularized build and command framework that makes it easy to expose new functionality through GUI and CLI