Perform Multiple Image Segmentation Techniques On High Resolution Image Using eCognition

Remote sensing & Digital Image Processing offers a wide variety of digital imagery that covers most of the Earth’s surface. This up-to-date image data is a tool for producing accurate land cover maps. To maximize the benefit of such data, automatic and efficient classification methods are needed. To achieve this objective (for the past years), pixel-based classification has been extensively used.

Currently the prospects of a new classification concept, object-based classification, are being investigated. Recent studies have proven the advantages of the new concept over traditional classifiers. The new concept’s basic principle is to make use of important information (shape, texture and contextual information) that is present only in meaningful image objects and their mutual relationships.

What is Image Segmentation?  In remote sensing, the process of image segmentation is defined as: “the search for homogeneous regions in an image and later the classification of these regions”. Available approaches can be grouped into three categories point based (e.g. grey-level threshold), edge-based (e.g. edge detection techniques) and region-based (e.g. split and merge). In the region-based category, image objects are generated according to a certain homogeneity criterion.

eCognition offers a relatively new segmentation technique called Multiresolution Segmentation (MS). Because MS is a bottom-up region-merging technique, it is regarded as a region-based algorithm. MS starts by considering each pixel as a separate object. Subsequently, pairs of image objects are merged to form bigger segments.

Image segmentation in general is defined as a process of partitioning an image into homogeneous groups such that each region is homogeneous but the union of no two adjacent regions is homogeneous (Pal and Pal, 1993).

Methodology

1.  Open the ecognition software.Select Rule Based Method on right. 

2.  Click on “create new project” option.

3.  Import the “Study area image” from the workspace.

4.  Click on the layers and rename each layer. Layer 1 as blue, layer 2 as green, and 3rd as red and 4th NIR.

5.   Go to “Process” option on the top toolbar and open “ process tree” All the operations in ecognition are executed through the “ process tree”

6.   Right click in “process tree” window and open the option “Append new”. Uncheck the “Automatic option” and type the name “Segmentation” and click OK.

7.  The parent class “segmentation” has been added. Now click the ‘segmentation” option and select the “Insert child” option to add sub classes.

8.  Click the first segmentation algorithm” chess board” and select the execute option.

9.  Execute the next algorithm “ Quad tree segmentation”

10. Execute Multi Resolution segmentation Scale 70 value -average size of objects Shape (0.5) weight -Geometric form of the objects Compactness (0.5) Value- The higher the value, the more compact image objects maybe.

11.  Execute Spectral difference segmentation with Maximum spectral difference value 15.

Flowchart

Quad Tree Segmentation

Multi Resolution Segmentation 

Conclusion 

In today’s tutorial a variety of Image segmentation techniques were applied.

First was Chess board segmentation which divides the image into equal sized squares. In chessboard segmentation the problem is that it only divide the picture into equal size and does not divide the picture on similarity base so various objects are mixed and can’t be differentiated, also they have no well-defined boundaries in segments. It would be better if we would have regular sized and properly shaped parcels in an image of a planned settlement etc.

Second was Quad tree segmentation that divides the image into different sized square objects. Quad tree segments the image on the basis of similarity of spectral response. The region in the image where the spectral response would be more similar so larger segments will be formed and in the region of dissimilarity segments are smaller in size.

Third was Multiresolution segmentation algorithm consecutively merges pixels or existing image objects.  Multiresolution segmentation is an optimization procedure which, for a given number of image objects, minimizes the average heterogeneity and maximizes their respective homogeneity. The segmentation procedure starts with single image objects of 1 (one) pixel size and merges them in several loops iteratively in pairs to larger units.

Fourth was Spectral difference segmentation, this is a merging algorithm where neighboring objects with a spectral mean below the threshold given (maximum spectral difference) will be merged to produce the final objects. To use this segmentation algorithm we are required to already have a segmentation (level) in place, we cannot create a new level using this algorithm. The objects are clearly segmented. Clear segments are formed between houses and their shadows. Vegetation, roads, houses and shadows are clearly segmented using this method.

References:

1.  K. Leukert, A. Darwish and W. Reinhardt, “Urban Land-cover Classification: An Object-based   Perspective,” 2nd Joint Workshop on Remote Sensing and Data Fusion over Urban Areas, Berlin,     May 2003.
2.  P. Mather, Computer processing of remotely-sensed images, Chichester: Wiley, 1999. [3] B.   Jähne,    Digitale Bildverar beitung, Berlin: Springer, 1993.
3.  R. Haralick, L. Shapiro, Computer and robot vision, Vol. 1, Reading, Mass.: Addison-Wesley, 1992.
4.  Definiens imaging, eCognition user guide [Online]. Available: http://www.definiens-   imaging.com/documents/index.htm, April 2003.
5.  NIMA, Geospatial Standards and Specifications [Online]. Available:   http://164.214.2.59/publications/specs/index.html, April 2003.
6.  Lillesand, T., Kiefer, R., “Remote Sensing and Image Interpretation”, New York, USA: John      Wiley & Sons, 2000.