Supervised classification is a threestep process: 1) Delineate training polygons of areas of known cover types, 2) Extraction of pixel statistics within each training poygon, 3) Use of maximum likelihood classifier with your pixel statistics to predict the class membership of every pixel in your original raster. In Arcmap we use the editor to create our training polygons, then the Create Signatures geoprocessing tool to extract pixel statistics within each training polygon, we then classify the original raster using the Maximum Likelihood Classification geoprocessing tool.