Abstract
Observers detected cylindrical features superimposed at known (prespecified) or unknown (random) locations within uniform circular areas (pedestals) on noisy computed-tomographic images. The experimental conditions varied the feature's contrast and size (3.0, 7.0, and 16.1 mm in diameter) for large (63 mm) pedestals, and varied the pedestal’s size from 9.8 to 63 mm for a fixed (7.0 mm) feature. As the pedestal increased in size, the feature’s detectability (measured by ROC analysis) increased to a constant level for all six observers, whether the feature’s location was known or random. There was no decrease in detectability for randomly located features, although the ratio of pedestal-to-feature areas increased by more than a factor of 10. Independent of the feature’s size, its detectability was predictable from the signal-to-noise ratio (SNR), as calculated for its matched filter at a known location. Detectability was proportional to the SNR when features were presented in specified locations. Larger values of SNR (higher feature contrast) were required to achieve comparable levels of detectability when the features were randomly located, but the required increase in the SNR did not depend on the relative areas of pedestal and feature. These data suggest that observers’ detection of features in random locations is relatively more efficient than the detection of known-location features.
© 1985 Optical Society of America
PDF ArticleMore Like This
WILLIAM H. STEIER and RAJ K. SHORI
FR3 Conference on Lasers and Electro-Optics (CLEO:S&I) 1985
Nesim Halyo and Richard W. Samms
ThD1 Machine Vision (MV) 1985
Friedrich O. Huck, Carl L. Fales, and Kathryn Stacy
ThC5 Machine Vision (MV) 1985