A constant-false-alarm-rate algorithm

Abstract The constant-false-alarm-rate (CFAR) detection algorithm is used for the detection of an optical target in an image dominated by optical clutter. The algorithm can be used for many aerial images when a clutter-subtraction technique is incorporated. To approximate the assumption of a constant covariance matrix, the digital image scene is partitioned into subimages and the CFAR algorithm is applied to the subimages. For each subimage, local means must be computed; the “best” local mean is the one that minimizes the third moment. The clutter subtraction technique leads to a mathematically tractable algorithm based on hypotheses testing. A test statistic and a threshold level must be computed. The value of the test statistic is subimage-dependent and is compared with the detection threshold which is chosen to specify a performance level for the test. A computationally efficient and stable implementation of the CFAR algorithm is given which may use either the Cholesky decomposition or the QR decomposition with a rearrangement of the computations.