Radial processing of correlated and uncorrelated image data

In this dissertation, the edge detection problem is investigated for two-dimensional images of correlated and uncorrelated data. The methodologies presented are based on ANOVA techniques which have been shown, in the past, to be robust and uniformly most powerful in the presence of independent Gaussian noise.;In most of the edge detection techniques, assumption is made that the source of errors is an additive independent Gaussian noise. However, the image may be of correlated data with unknown statistics.;In this work, the ANOVA mask scans the image radially at certain angle, initiating the process from the inside of the image to be processed. The radial processing combines the good qualities of standard ANOVA and gradient techniques.;The performance of ANOVA and adjusted ANOVA using the F-statistic as the decision (threshold) function, combined with the test on contrasts is examined through simulation on three digital image models: signal in independent Gaussian noise, autoregressive model and signal in Markov correlated noise. The performances are compared to the standard ANOVA techniques, in which the mask scans the image horizontally or vertically.;The results in the three cases illustrate the efficiency and robustness of the proposed procedures.