Comparative Evaluations of the RADC/Hsu Texture Measurement System with Perceptual Analyses
Abstract
Earlier it was determined that the essential information for terrain discrimination was contained in three tone-texture variables: mean density, first and second neighbor contrasts. In this study, this fundamental concept was successfully sustained using a Monte Carlo technique. To determine the relationships between the RADC/Hsu texture measurement system and human perceptual judgments, four population maps with 20 subjects and four terrain types with 40 subjects were analyzed. The first analysis (20 subjects) yielded a clear-cut tone and texture 'perception model' with a multi-dimensional scaling technique. The second analysis (40 subjects), according to a minimum stress criterion (free-running), yielded a less-interpretable tone and 'texture plus structure' perception model. A fixed model using computed tone and texture parameters, however, gave a more satisfactory and interpretable result. Accordingly, 50% of the subjects indicate a near-perfect fit, another 25% have a moderate fit and the rest (25%) belong to lack-of-fit and no-fit categories. The man-machine interaction pattern in these models reveal that the machine classifier weighted the tone parameter heavier than the texture parameter by a factor of 1.5, whereas the human subjects displayed interesting individual differences as to how they weighted these two dimensions.
Document Details
- Document Type
- Technical Report
- Publication Date
- Oct 01, 1979
- Accession Number
- ADA079278
Entities
People
- Richard G. Burright
- Shin-yi Hsu