The purpose of image classification is to produce meaningful land cover maps by the identification of individual pixels or groups of pixels with similar spectral responses (spectral signatures) to incident and/or emitted radiation. Pixels or groups with similar values represent different materials or classes. The actual values associated with each pixel are analyzed mathematically using computer driven algorithms. These algorithms attempt to determine the uniqueness of classes and to cluster similar pixels and groups of pixels into classes. The computer automatically creates image classes and partitions the data through a clustering process. The clustering algorithms are bundled with all commercially available image processing software programmes.
An alternate method of image classification is the unsupervised classification, which examines a large number of unknown pixels and divides them into a number of user-defined classed based on natural groupings present in the image values. In the image dataset, pixels are clustered statistically without any user defined classes. Unlike supervised classification, the unsupervised classification does not require the analyst to specify training areas. The basic premise of unsupervised classification is that any given land cover type will have similar DN values of pixels or similar gray levels, whereas data in different classes will be comparatively well separated.
The result of unsupervised classification will yield results which based on natural groupings of the image values. The identity of the spectral class will not be initially known, but can be determined by comparing the classified data with some reference data such as large scale topographic maps or imageries, or by site visits. Thus, in the supervised classification approach, useful information categories are defined and then their spectral separability is examined; in the unsupervised approach the computer determines spectrally separable class, and then defines their information value. Thus, although the unsupervised classification approach requires no user input to create the classified image, the output tends to require a great deal of post classification operations to make the results more meaningful.
The unsupervised approach to image classification is becoming increasingly popular in agencies involved in long term GIS database maintenance. This is prompted by the availability of improved algorithms that use clustering procedures that are extremely fast and require little human input by way of operational parameters. The other reason is the availability of an overabundance of satellite data which demands automation and lesser dependence on supervision on the part of the analyst. Thus it is becoming possible to train GIS analysts with only a general familiarity with remote sensing to undertake classifications that meet typical map accuracy standards. With suitable ground truth accuracy assessment procedures, this approach to image classification and information extraction can provide a remarkably rapid means of producing quality land cover and thematic maps on a continuing basis.
This website is hosted by
Department of Geology
Aligarh Muslim University, Aligarh - 202 002 (India)