VISUALIZATION OF THE RELIABILITY IN CLASSIFIED REMOTELY-SENSED IMAGES

Much research has been concentrated on studies of the nature and cause of unreliability in classifying remotely sensed images, particularly those derived from the Landsat series of satellites. This paper presen ts a methodology to allow a variety of measures of that accuracy (Over all Accuracy, User and Producer Accuracy, and Classification Accuracy) to be embedded in the display of the classified image. The principle is that, as the image is displayed on a computer screen, pixels are ra ndomly selected and the cover type at that location is re-displayed ac cording to algorithms listed here. The resulting displays are fundamen tally dynamic, although static images are used to illustrate the paper . ft is believed that the method gives a new dimension to the visualiz ation of classified images, and makes appropriate use of computer reso urces.