Object-based Image Analysis

The course ‚Object-based Image Analysis‘ was a deep dive into the topics of remote sensing and image analysis, providing advanced insights into the technique of object-based image analysis (OBIA).

In difference to other traditional image analysis approaches, OBIA does not work on the pixel level of the image but instead creates so called image objects first. This image segmentation approach is intended to divide the image into areas with similar (spectral) properties on a defined scale, such as individual building footprints or forest patches. This eliminates the risk of salt-and-pepper effects found in pixel based classifications, especially in high resolution imagery, as extreme values have a lower impact on the final classification results. In the second crucial step, the image objects are classified based on the selected classification algorithm.

OBIA also allows the user to incorporate additional information into the classification process, such as the image object’s geometry and statistical properties, the relation to other neighbouring image objects and scale levels or even external vector layers imported from a GIS. This inclusion of GIS functionalities and rule-based class modelling presents a opportunity to complement the classification process and improve the final classification results. It also allows for the incorporation of semantic classes into the classification, such as the classification of orchards, a land use type cannot be classified based on the spectral information of the trees alone.

Chapters of the course did include fundamentals of image interpretation and perception, basic concepts of hierarchy theory and knowledge representation, as well as more advanced concepts of image segmentation and (object-based) classification including class modelling. The final two chapters were focused on the application of accuracy assessments (including object validity) for OBIA classifications and further areas of application for non-image data.

Key objectives were to acquire an overall understanding of OBIA as an advanced image understanding theory, the application of spatial concepts in image analysis including concepts of geometry and neighbourhood, as well as gaining practical experience inside the software environment of eCognition Developer for the different parts of the modelling workflow.

Every chapter did include a practical exercise part besides the lecture itself making it easy to apply the gained knowledge in a intuitive way. The course ended with a final exam in which the concepts and techniques learned during the course were tested by means of a practical example.

Beitrag veröffentlicht