Accession Number : AD0759505

Title :   Texture Tone Study. Classification Experiments.

Descriptive Note : Interim technical rept. no. 4,

Corporate Author : KANSAS UNIV/CENTER FOR RESEARCH INC LAWRENCE REMOTE SENSING LAB

Personal Author(s) : Dinstein,Its' hak ; Haralick,Robert M. ; Shanmugam,S. K. ; Goel,D.

Report Date : 31 DEC 1972

Pagination or Media Count : 207

Abstract : Four aerial photographic image data sets were classified on the basis of a large class of quickly computable textural features. When the most appropriate features and decision rule were selected, identification accuracy on the order of 75 per cent was obtained for 9 to 11 terrain categories. Conclusions drawn from these experiments suggest: That the most powerful features are the entropy and inverse difference features measured at distance 1 and at distance 1/10th the length of the image side; That the class of quickly computable textural features needs to be supplemented by tonal and context features in order for better identification to be obtained. This second conclusion is not to be unexpected since a photointerpreter who tries to make interpretations on the basis of a 1/8 inch x 1/8 inch squared specially processed for high contrast on a 9 inch x 9 inch 1:20,000 aerial photograph does not do any better than 75 per cent correct identification as previously reported. (Author)

Descriptors :   (*PHOTOGRAPHIC IMAGES, PHOTOINTERPRETATION), (*DATA PROCESSING, *PATTERN RECOGNITION), PHOTOGRAPHIC TEXTURE, PHOTOGRAPHIC TONE, AERIAL PHOTOGRAPHY, DECISION THEORY, MINIMAX TECHNIQUE, COMPUTER PROGRAMS, ENTROPY, IDENTIFICATION SYSTEMS, INFORMATION THEORY

Subject Categories : Cartography and Aerial Photography
      Computer Programming and Software
      Cybernetics

Distribution Statement : APPROVED FOR PUBLIC RELEASE