Image Classification

Classification is the process of sorting pixels into a finite number of individual classes, or categories of data, based on their data file values. If a pixel satisfies a certain set of criteria, then the pixel is assigned to the class that corresponds to that criteria.

For the first part of the classification process, the computer system must be trained to recognize patterns in the data. Training is the process of defining the criteria by which these patterns are recognized. The result of training is a set of signatures, which are criteria for a set of proposed classes.

There are two ways to classify pixels into different categories: 

Supervised unsupervised

Classification tools are available on the Raster tab.

See Classification Process in HexGeoWiki for more information about classifying a raster file.

To learn how to perform and evaluate image classification, see Classification Workflows Contents.

Supervised Classification

Supervised Classification uses existing region or class statistics to classify an image. The statistics might be based on sample training regions drawn by hand, using Annotation Tools or converted from a vector image using Regions menu options. Or they could be output from the Unsupervised Classification utility or a previous supervised classification.

The regions you use for supervised classification must already have statistics calculated for them. Statistics are calculated automatically for class regions created by the unsupervised classification utility but for training regions (created using Edit/Create Regions) statistics must be calculated using the Calculate Statistics option on the Process menu for the particular class or region.

Specify images and bands

1. Check that statistics have been calculated for the image.
2. In View menu select Statistics. Select Show Statistics. The Statistics Report dialog opens. Click OK. The Display Dataset Statistics dialog opens.
   If the Display Dataset Statistics dialog is empty, you must calculate the image statistics. See Regions and Statistics for information.
3. From the Process menu select Classification. Select Supervised Classification. The Supervised Classification dialog opens.
   
4. Set the following as required:
• Input Dataset: The image to be classified. This image is not affected during classification.
• Input Bands: List of bands you want to include in the calculation; for example: 1,4,7 or 2-4, 5. For Landsat TM data, you would typically use bands 1-5 and 7. These may be typed or selected from the band chooser.
• To select from the band chooser, click the Input Bands chooser menu button. A list of the bands in the input image appears. Drag to highlight consecutive bands. Select or unselect individual bands by pressing Ctrl‑click. Default entry is ‘All’.
• Output Dataset: New classified image. It defaults to the same name as the Input Dataset, with _class added to the end of the file name; this can be changed if required.
• Band 1 of the output image has the same form as an image created using the unsupervised classification utility.

Unsupervised Classification

Unsupervised Classification is carried out using little or no information about possible classes and automatically identifying clusters of similar data.

The ISOCLASS algorithm, used for unsupervised classification, groups data automatically, recalculating class means, and merging and splitting classes as required. Several parameters control processing and completing the classification; these default values can be changed if required. A full set of statistics for the classes is automatically calculated after classification is completed.

ISOCLASS is the clustering analysis process used by ER Mapper for unsupervised classification.

To carry out a simple classification, just specify the input and output images. ER Mapper uses the default parameters to carry out the classification. To perform a more complex classification, you can set the parameters yourself.