Automatic Terrain Classification From Photography

As the acquisition of aerial photographic imagery continues to grow at a rapid rate, automatic data extraction becomes more and more desirable. To this end, an image processing system is being developed to computerize terrain classification. A laser scanner has been used to digitize Apollo 9 imagery of the Salton Sea area. A signature algorithm has been specified allowing the assignment of small image blocks to specific terrain categories. Results of automatic classification and manual photointerpretation are compared. Categories considered are cultivated terrain, uncultivated terrain, cloud covered terrain, and water. Assignment is based on a single black and white transparency in each case. The assignments are displayed as overwritten alphabetic annotation on the original image. Problems are discussed.