Evaluation of the Two-Dimensional Fourier Transform Applied to Acoustic Image Pattern Measurements.

摘要

The objective of this experimental study was to evaluate the use of the two-dimensional Fourier transform as an aid in extracting spatial frequency components in acoustic data images by means of direct operator observation. More specifically, the study addressed the question of the relative gain in precision of measurement which might be realized using the two-dimensional transform as compared to making these measurements directly from the original image. Of particular interest was the amount of gain at the weaker signal strengths typical of the class of processed acoustic data under investigation. As the signal power in the image decreases relative to the noise power, the probability increases that energy concentrations resulting from random noise patterning will appear in the transformed image. It is possible, therefore, that the human operator working with the original image alone may be able to separate signal related patterning from spurious noise patterning as well as can be achieved by using the transform. The experiment reported herein was designed to provide a definitive test of this hypothesis. Results showed that at high SNR levels the transform domain image representation was clearly superior to the original image as a mode for systematic measurement and ranking of periodic spatial features in the image. As the SNR level was reduced the accuracy of measurement in the transform domain fell off more rapidly than the original image domain, resulting in a convergence of the performance curves under the two modes. With the SNR level reduced so that the signal-related patterning was barely perceptible, measurement of periodic spatial features in the transform domain was no better than in the original image domain. (Author)