This is the second of the six lectures being devoted to the components that make up digital image analysis (the first lecture--the introduction--was given on Jan 29). In this lecture I will review unsupervised classification, which is often the first step taken in any image interpretation exercise since it provides the analyst with an overview of the information content present in the images.

The Biodiversity Informatics Facility provides an overview of image interpretation and classification that is well worth reading. The Canadian Centre for Remote Sensing also provides an overview of image classification and analysis. The FAO has produced a comprehensive guide to Land Cover Classification concepts (related to a free software program they also supply).

A UNESCO project--you can download free remote sensing software (BILKO) as well as relevant remote sensing tutorials (on, for example, unsupervised and supervised classification).

An animation of k-means clustering (which is similar to the iso cluster routine), which illustrates how the groups are formed (around the 'seeds', the number of which reflects the number of groups you specifiy when you initiate the routine).

Text: Chapter 8: Digital Image Interpretation

ESRI (ArcMap) provides some general overviews of image classification:

• A discussion of multivariate classification by ESRI, of which unsupervised and supervised classification are the two main types.
• An explanation of image classification along with an example of a supervised classification.
• An overview of the entire image classification process (well worth reviewing).
• A description of the Iso Cluster routine implemented in ArcMap.

A commonly used classification scheme has been the Anderson Classification System - A land cover/land use classification system developed for use with remote sensing systems in the 1970's, adopted for the USGS-NPS Vegetation Mapping Program to map cultural land cover (Anderson et al. 1976).

Note: This is not the complete Anderson Level II Classification--vegetation classes have been omitted (obtained from USGS).