What is Genie Pro?

Genie Pro™ is an easy to use, general purpose, interactive, adaptive, automated feature extraction (AFE) software tool. Genie Pro works by automatically developing image processing algorithms for detecting features of interest based on user examples of those features. Genie Pro accelerates the feature extraction process overcoming the time-consuming procedures of manual feature identification by using cutting edge learning techniques to recognize your features of interest.

Genie Pro’s core concept is the use of machine learning technology to learn or “discover” the key signatures of features of interest. Specifically, it uses an evolutionary algorithm, a specific type of machine learning, to evolve an image processing program that closely duplicates the training results provided by the user. The resulting image processing program can be applied to much larger sets of image data, with the results used to label regions of similar content, known as classification, or to identify a particular type of feature or object in large amounts of image data, known as search.

The power of Genie Pro over most other classification and search techniques is that it can use both spatial and spectral processing techniques to develop its program.  Spatial relationships such as texture, proximity, or shape are added to the spectral information as additional dimensions of the attribute vector for each pixel. This is important because many classification and search tasks have both spatial and spectral components to their signatures.

Genie Pro is designed for pixel-by-pixel classification and can be used to label pixels in multiple imagery types for, among other things, terrain classification and crop identification. It can be used to locate distinct objects such as planes or vehicles and for extended objects like roads or buildings. It can be used in single or multi-class classifications.  

Major Components

  1. A Graphical User Interface (GUI) for providing training data and inspecting results.
  2. An adaptive spectral/textural image processing stage for learning the attributes of pixels containing the feature of interest.
  3. An adaptive morphological image processing stage for refining the results of the spectral/textural image processing stage.
  4. A supervised statistical classifier (regularized maximum likelihood Gaussian classifier) that uses the training data from the GUI and the attributes found by the spectral/textural and morphological image processing stages to determine the decision rules for extracting the feature of interest.
  5. An adaptive thresholding algorithm that takes the result of the statistical classifier and produces a raster map of the feature of interest in the image.
  6. A vectorization algorithm designed to smoothly delineate raster regions containing the features of interest. This enables export of Genie Pro’sresults to standard GIS applications.

Genie Pro Components

What's Under the Hood?

Image Processing Operators

Genie Pro uses a library of basic image processing operations and a grammar-guided evolutionary algorithm to build non-linear spatial-spectral image processing chains that define a pixel attribute space in which a traditional supervised statistical classifier can learn the decision rules for each class. Genie Pro’s spatial processing operators include both textural and morphological processing operations. Textural operators calculate neighborhood properties, such as statistics, and morphological operators apply shape-based filters to the data. The resulting attribute space can provide interesting visualizations of the data that can be analyzed to investigate the signatures of the feature(s) of interest.

Grammar Files

Genie Pro uses Grammar Files to guide how it builds image processing solutions. You can think of a Grammar File as a template for how to build an image processing chain. Each attribute in a Genie Pro solution consists of one image processing chain. The Grammar File consists of a set of rules that determine the types of image processing operators that can be used, the input parameters that are needed for each operator, the way in which the operators can be connected into a processing chain, the probability with which the operators are randomly selected, and the probability with which input parameters are chosen.

A number of Grammar Files come with Genie Pro. These have been designed to have specific characteristics that may be useful for certain image sources or situations. The Standard Grammar file was developed with a good balance of solution complexity, speed and accuracy.

The Learning Engines

Machine learning for remote sensing requires a method to automate feature extraction. It needs to be able to learn to create a representation of the feature(s) and background. It then must have a classifier to separate the feature from the background or other classes. There are many ways to choose this representation through spatial and spectral pre-processing of data, and many possible classifier algorithms (classical statistics, decision trees, artificial neural networks, hidden Markov models, clustering, support vector machines, etc). Genie Pro uses one of two learning engines: (1) the Ifrit learning engine, which uses an evolutionary algorithm for image chain discovery and the Fisher linear discriminant algorithm for classification, and (2) the Djinn learning engine, which uses a user-adjustable fixed set of feature image processing chains and one of two algorithms for classification.


The Solution Graph Visualizer is a powerful tool for understanding both how Genie Pro works and what it is using to generate your solutions. Visualization of your Genie Pro solutions allows viewing a solution as a set of ordered graphs, where an "attribute image" can be generated and displayed for each node of the graph.

Vectorization Tool

The vectorization tool allows the user to create standard ESRI shapefiles for any or all of the classes created in Genie Pro. These shapefiles can then be exported to any GIS software for additional analysis.

Command-Line Control

For more advanced users a command-line capability is available for Genie Pro. The command-line utility allows you to run batch training or batch solutions on large sets of images.

Feature Descriptions

Comprehensive image file format support: Genie Pro ingests over 30 commonly used geospatial image file formats, including NITF 2.0/2.1, GeoTIFF, and Erdas Imagine.

User friendly training markup tools: “Ground truth” training data is supplied to the system via an intuitive paintbrush-style interface.

Ifrit adaptive feature extraction engine: The Ifrit engine is at the heart of Genie Pro, and is responsible for deriving robust and accurate feature extraction solutions from user-supplied training data.

Djinn adaptive feature extraction engine: The Djinn engine is based around a different learning technology to Ifrit, and is often faster and easier to use on simpler problems.

Outline vectorization tool: Converts the raw raster results files produced by Genie Pro solutions into editable vector overlays.

Centroid vectorization tool: Vectorizes the center of distinct classified regions within the results layer rather than the perimeter.

Object Detection: An alternative learning mode geared toward object detection.

Error analysis tools: Provides an error assessment of the solution against the training data. It can also be used to assess a solution against ground truth data.

Raster files results export: Results layers are exported in a TIF format.

Shapefile results export: Vectorized results (both outline and centroid layers) can be exported to standard shapefiles that can be read by standard GIS software.

Command line tools for batch processing: For automatic processing of large datasets, most of the functionality of Genie Pro can be accessed via command line tools. Command line tools were added in version 2.0. and included tasks such as generating object detection ROC curves and importing training data from external sources.

Image viewer pan, zoom and band selection controls: Genie Pro includes a sophisticated image viewer that allows the user to navigate around very large images easily, and to select band combinations for viewing.

Results threshold sliders: Feature detection and background rejection thresholds may be adjusted once a result layer has been produced. This adjusts the threshold values of the classifier between the various classes.

Now our Tutorials are on YouTube!

Visit our Genie Pro YouTube Channel. Add your comments and make requests for new information you'd like to hear about.

Subscribe to us on YouTube

Genie Pro Channel

Want to See Genie Pro in Action?

We have added new features to enhance your solutions. We even took your input into consideration and added several user requested features as well. View some of our case studies and see Genie Pro in action.

Read more

Observera, Inc.