A constant-false-alarm-rate algorithm (Q1194522): Difference between revisions

Given a stationary scene, the problem is to determine whether or not a particular optical target is located in the scene. Let an \(m*N\) data matrix \(X\) represent the digital image scene to be tested and let \(\hat X=\omega^{-2}[X\circledast W]\), where \(W\) is an \(\omega\) by \(\omega\) all ones matrix and \(\circledast\) denotes discrete convolution. The value of a typical pixel element of \(\hat X\) is the average of the \(\omega^ 2\) pixel elements that are in the square neighborhood (window of size \(\omega\)) of the corresponding pixel element of \(X\). Then the hypotheses which the detector algorithm must test is \(H_ 0: X=X_ 0\) (clutter only) against \(H_ A: X=X_ 0+bs^ T_ 0\) (clutter plus signal). A generalized likelihood ratio (GLR) test is derived and its computational properties are studied. To reach an as much as precise solution, the Cholesky (\(QR\) respectively) decomposition of \(X\) is used. The authors implemented the method in the C language and present their experience with the numerical results obtained for simulated data set in the last section.