Image Classification toolbar introduced at ArcGIS 10

A new toolbar, Image Classification, has been introduced at ArcGIS 10 to make image classification tasks both faster and easier (to use this toolbar, you need the Spatial Analyst extension). Expressed simply, the image classification process converts multiband raster imagery into a single-band raster with a number of classes, which you can then use to make thematic maps or for further analysis. Example applications for image classification include landcover mapping and landuse change detection.

The following images represent an example of landuse classification. The first image is a false-color infrared image (from the Landsat 7 TM sensor) of the northern Cincinnati area, and the second image is the result of landuse classification using the Image Classification toolbar.

Raw satellite image

Landuse classification result


Main functionality


The toolbar offers the following functionality:

* - Interactive tools for creating training samples
* - A window to manage your classes and training samples
* - Three training sample evaluation windows that allow you to evaluate classes as histograms, scatterplots, or by their statistics
* - Easy access to the classification tools from the Spatial Analyst Multivariate toolset

There are two main ways to classify a multiband raster image; supervised and unsupervised classification. The toolbar is primarily intended for supervised classification, which employs your knowledge of the study area to achieve good results. In a supervised classification, an image is classified using polygons that represent distinct sample areas of the different land use types to be classified (training samples). In previous releases, training samples were collected by first having to create a polygon feature class and then having to edit the feature class using the Editor toolbar. In ArcGIS 10, this task has been made much easier by interactive creation and editing of training samples with the Image Classification toolbar.

Performing supervised classification with the Image Classification toolbar

The basic functionality of the Image Classification toolbar is illustrated in the following workflow. To perform a supervised classification, generally you need to go through three steps: collecting the training samples, generating a signature file, and executing the classification tool.

1. Collecting the training samples

To collect training samples, you need to find areas with relatively uniform appearance on the image, and then use one of the drawing tools from the Image Classification toolbar to draw graphics to enclose them. For each graphic that you draw in the display, a class is created in the training sample manager. You can edit the class name, class value, and change its color.



You can also view the value distribution of the training samples. To do so, select the classes in the training sample manager and click any of the three evaluation tools.

2. Generating a signature file


Once the training samples are ready, you can create the signature file. To do so, open the training sample manager and click the Create signature file button. In the resulting Save As file browser dialog box, pick a path and name for the signature file, then click Save.



3. Executing the Maximum Likelihood Classification tool

The final step is executing the Maximum Likelihood Classification tool is to classify the image. This tool is accessible through the Classification drop-down list on the Image Classification toolbar. To execute this tool, input the image to be classified as Input raster bands, and the signature file you created in step 2 as the Input signature file.

4. Obtaining final results

The classified image from step 3 often needs further processing to clean up the random noises and small isolated regions to improve the quality of the output. This process is known as post-classification processing, which is explained in full details in the desktop help.

Source: ESRI
Previous
Next Post »