ILMSimage 2.4 Feature Calculation
ILMSImage Feature Selection assigns attributes or features to each cell in the index. All classification work depends on cells and cell features. The features can be calculated directly from the supplied image data or added from external raster layers.
Feature Selection shows a complete list of all features, that ILMSimage can derive directly from the loaded image layers. To select features for further processing they must be checked by clicking the check box. The first four items correspond to the shape of the cells, the following items make all color bands in the loaded image layers available and the last five items comprise texture and deviation features. Add Image Data opens new image bands to select features calculated elsewhere.
- Cell Size selects the area covered by the cell as feature. All Cell Features are measured in pixels.
- Cell Perimeter selects the perimeter of the cell measured in pixel borders. In ordinary cases Cell Size and Cell Perimeter are highly correlated.
- Cell Length selects the diagonal measure of the smallest rectangle that completely surrounds the cell
- Cell Width selects the width of the biggest square that can be enclosed by the cell
- Input Band selects the medium color density of all pixel values within one cell
- Contrast (linear) selects the standard contrast calculation from the layer with the highest spatial resolution. If all layers show the same resolution or more than one layer has a higher resolution than others, the contrast is calculated using the first principal component.
- Contrast (normalized) selects the standard contrast calculation using the modulation of the pixel densities.
- Contrast (invers, IDM) selects the Inverse Difference Moment (Haralick et al. (1973), S. 619) for contrast calculation.
- Standard Deviation selects the standard deviation of all pixels within the cell as feature
- Variance selects the variance, i.e. the square of the standard deviation as feature.
- Define Formula is deactivated.
[Add Image Data] opens the system file selection dialog and adds one or more new raster layers to the selection list. According to image selection at the begin of the project project settings ILMSimage stores a copy of the selected data in the project folder and converts it to a suitable format.
Show Feature Image creates a multiband image layer that represents all selected cell features and shows it on the QuantumGIS canvas. The feature image will normalize the selected features in the same way the classification routines will do, so the information transferred to the classification process can be visualized. Users unfamiliar with object orientated approaches should use this opportunity to get an impression about the significance of the different figures in an attribute table.
The panel Control provides the option Show Fearures Table to control the result of the feature calculation as a formatted table. If the cell index is transformated to a shape file with the option Add Features to Cell Index Table the same table will be available as attribute table of the cell index shapefile.
Features may be recalculated without changing the cell index but all subsequent steps must be repeated with the new feature values.
Using Additional Data Layers
In addition to the mentioned standard attributes, external data layers can be integrated into the process of feature calculation. For the integration basically every available dataset in raster form is suitable. In the integration within every layer and for every cell a mean value is calculated and finally added to the feature table. This means that features from additional data layers are treated equivalent to the raster layers selected at the begin of the project, the only difference being that additional data layers do not influence cell creation. This function is therefore suitable for integrating additional multispectral levels (or combinations of these) which are required as attributes for additional analysis but not for a derivation of analysis areas, i.e. cells, into the work process. A simple example is the Normalized Vegetation Difference Index, NDVI which is a measure for vegetation activity. It can be calculated using two channels which are in the visual field and in the close infrared field of the electromagnetic spectrum. As additional data layer a pixel-based NDVI picture is available. The following image shows areas of high vegetation activity in bright and areas of low activity in dark grey levels.
This dataset can now be defined as an external data layer - for this purpose it does not have to be loaded in QuantumGIS; it is sufficient to know the corresponding path.
Accomplishing Feature Calculation
After clicking the [Run] button the selected features are calculated and a small window with a progress bar will show up and provide a message about the ongoing operation. All messages of the current run are listed in the Control panel and stored in a logfile using the project name for further use.
If the Show Feature Image option is active, the feature layer will be loaded to the QuantumGIS canvas. All features are accessible in a table view from the Control panel. If feature calculation has finished sucessfully, ILMSimage will show the Cluster panel.