InterIMAGE

In the following some basic characteristics of InterIMAGE will be outlined, so that the reader can better understand how it can be used and what we plan to do with the system within TOLOMEO.

InterIMAGE is an open-source system (GPL Licence), developed in C++, using the Qt and TerraLib libraries.

InteIMAGE is actually a generic image analysis tool that can be regarded as a knowledge- and object-based system. The system ultimately provides tools for the users to create the so called “interpretation knowledge models”, and the basic components of the models are:

(i) a semantic network – which is rather a tree whose nodes refer to the classes of objects expected to be found in the scene to be analyzed (Figure 1).


Figure 1. Example of a semantic network in InterIMAGE

(ii) operators – which are executable programs that can be associated to the nodes of the semantic network and that are automatically started by InterIMAGE’s control process during the execution of the interpretation knowledge model. Operators can vary in scope and complexity – they can implement generic analysis processes, such as image segmentation or thresholding, or complex algorithms for the extraction of particular object classes (e.g. to find rivers or red roof houses).

(iii) decision rules – which are user defined rules, also associated to nodes of the semantic network, that basically processes the outcome of operators. Decision rules may be set to be executed right after the execution of an operator, for instance, there may be a node “Pool” in the semantic network, to which a generic segmentation operator is associated, after the execution of the segmentation operator, the decision rule can be set for selecting only the rectangular blue segments (Figure 2a). A decision rule can also be used to resolve eventual conflicts among overlapping objects associated to different nodes (classes) of the semantic network based on the objects membership values (Figure 2b).


Figure 2. Example of decision rules in InterIMAGE

In simple terms InterIMAGE’s interpretation process manages the execution of the operators and decision rules. Starting from the root node of the semantic network, it commands the execution of the operator and decision rule associated to the node, waits for the respective results and then executes the operators and decision rules associated to the child nodes (in parallel). This process (top-down) runs recursively until the leaf nodes are processed. After that, the interpretation process visits the nodes in the opposite direction (bottom-up), executing for the parent nodes decision rules defined to resolve eventual spatial overlaps among hypotheses associated to sibling nodes.

InterIMAGE is deployed together with a few generic operators. Such suite of generic operators currently consists of:

  • TA_Arithmetic – limiarization (thresholding) over an artificial band created through band arithmetic;
  • TA_NDVI_Segmenter – limiarization over NDVI;
  • TA_Region_Growing_Segmenter – region growing segmentation (SPRING’s algorithm);
  • TA_Baatz_Segmenter – region growing segmentation (eCognition’s algorithm);
  • TA_C45_Classifier – decision tree supervised classifier (C4.5 algorithm);
  • TA_ShapeFile_Import – import vector (shapefile) GIS layers; and
  • TA_ShapeFile_Intersect – intersect vector (shapefile) GIS layers.

It is very simple to add a new operator to the system – there is a simple protocol to be followed so that InterIMAGE will know how to call the operator’s executable and also know where to look for its results. Operators can be coded with virtually any computer language, since InterIMAGE will only interact with executable files.

The decision rules are different from operators in the sense that the environment for the definition of the rules, and the code of all functions that can compose a decision rule are parts of the system’s core, i.e., are programmed within the system. Some of these functions calculate object attributes. Currently approximately 65 object attributes can be calculated by the system.

Publications: 

COSTA, G.A.O.P.; FEITOSA, R.Q.; FONSECA, L.M.G.; OLIVEIRA, D.A.B.; FERREIRA, R.S.; CASTEJON, E.F. Knowledge-based Interpretation of Remote Sensing Data With the InterIMAGE System: Major Characteristics and Recent Developments. GEOBIA 2010. Gent, Belgium, 2010.

KUX, H.J.H.; NOVACK, T.; FERREIRA, R.; OLIVEIRA, D.A.B. Urban land cover classification using optical VHR data and the knowledge-based system InterIMAGE. In: GEOBIA 2010, 2010, Ghent, Belgiun. Proceedings of the GEOBIA 2010, 2010.

NOVACK, T.; KUX, H.J.H.; FEITOSA, R.Q.; COSTA, G.A.O.P. Per-block urban land use interpretation using VHR data and the knowledge-based system InterIMAGE. In: GEOBIA 2010, 2010, Ghent, Belgiun. Proceedings of the GEOBIA 2010, 2010.

COSTA, G.A.O.P.; PINHO, C.M.D.; FEITOSA, R.Q.; ALMEIDA, C.M.; KUX, H.J.H.; OLIVEIRA, D.A.B.; FONSECA, L.M.G. INTERIMAGE: An Open Source Platform for Automatic Image Interpretation. In: II Simpósio Brasileiro de Geomática e V Colóquio Brasileiro de Ciências Geodésicas, 2007, Presidente Prudente. Anais do II Simpósio Brasileiro de Geomática e V Colóquio Brasileiro de Ciências Geodésicas. Presidente Prudente, SP : UNESP - PP, 2007.

COSTA, G.A.O.P.; PINHO, C.M.D.; FEITOSA, R.Q.; ALMEIDA, C.M.; KUX, H.J.H.; OLIVEIRA, D.A.B. An Open Source Environment for Image Interpretation in Oil Industry Applications. In: Iberian-American Conference on Technology Innovation and Strategic Areas - 2007, 2007, Rio de Janeiro, RJ. Proceedings of the Iberian-American Conference on Technology Innovation and Strategic Areas - 2007. Rio de Janeiro, RJ : PUC-Rio, 2007.

The respective digital versions of the articles listed above can be downloaded from:

http://wiki.dpi.inpe.br/doku.php?id=interimage:publications