Use ERDAS IMAGINE Feature Space tools to interactively define feature space objects (AOIs) in the feature space image. A feature space image is simply a graph of the data file values of one band of data against the values of another band (often called a scatterplot). In ERDAS IMAGINE, a feature space image has the same data structure as a raster image; therefore, feature space images can be used with other ERDAS IMAGINE utilities, including zoom, color level slicing, virtual roam, Spatial Modeler, and Map Composer.
Example of a Feature Space Image
The transformation of a multilayer raster image into a feature space image is done by mapping the input pixel values to a position in the feature space image. This transformation defines only the pixel position in the feature space image. It does not define the pixel’s value.
The pixel values in the feature space image can be the accumulated frequency, which is calculated when the feature space image is defined. The pixel values can also be provided by a thematic raster layer of the same geometry as the source multilayer image. Mapping a thematic layer into a feature space image can be useful for evaluating the validity of the parametric and nonparametric decision boundaries of a classification (Kloer, 1994).
When you display a feature space image file ( .fsp.img) in a View, the colors reflect the density of points for both bands.
- Bright tones represent a high density.
- Dark tones represent a low density.
Create Nonparametric Signature
You can define a feature space object (AOI) in the feature space image and use it directly as a nonparametric signature. Since the Views for the feature space image and the image being classified are both linked to the Signature Editor, it is possible to mask AOIs from the image being classified to the feature space image, and vice versa. You can also directly link a cursor in the image View to the feature space View. These functions help determine a location for the AOI in the feature space image.
A single feature space image, but multiple AOIs, can be used to define the signature. This signature is taken within the feature space image, not the image being classified. The pixels in the image that correspond to the data file values in the signature (that is, feature space object) are assigned to that class.
One fundamental difference between using the feature space image to define a training sample and the other traditional methods is that it is a nonparametric signature. The decisions made in the classification process have no dependency on the statistics of the pixels. This helps improve classification accuracies for specific nonnormal classes, such as urban and exposed rock (Faust et al, 1991).
See Feature Space Images for more information.
Process for Defining a Feature Space Object
Process for Defining a Feature Space Object (for the above figure)
- Display the image file to be classified in a View. In Figure 186, the image bands are 3, 2, 1.
- Create feature space image from the image file being classified (layer 1 compared to layer 2).
- Draw an AOI (feature space object) around the study area in the feature space image. Once the AOI is created, it can be used as a signature.
- Use one of the decision rules to analyze each pixel in the image file being classified. The pixels containing the corresponding data file values are assigned to the feature space class.
Evaluate Feature Space Signatures
Using the Feature Space tools, it is also possible to use a feature space signature to generate a mask. Once it is defined as a mask, the pixels under the mask are identified in the image file and highlighted in the Viewer. The image displayed in the View must be the image from which the feature space image was created. This process helps you to visually analyze the correlations between various spectral bands to determine which combination of bands brings out the desired features in the image.
You can have as many feature space images with different band combinations as desired. Any polygon or rectangle in these feature space images can be used as a nonparametric signature. However, only one feature space image can be used per signature. The polygons in the feature space image can be easily modified or masked until the desired regions of the image have been identified.
Use the Feature Space tools in the Signature Editor to create a feature space image and mask the signature. Use the Insert Geometry tools to draw polygons.
Provide an accurate way to classify a class with a nonnormal distribution (for example, residential and urban).
Classification decision process allows overlap and unclassified pixels.
Certain features may be more visually identifiable in a feature space image.
Feature space image may be difficult to interpret.
Classification decision process is fast.