Considering complex scenes of log-normal sea clutter, we propose and analyze the automatic bilateral censoring and detection performances of the AML-CFAR (Approximate Maximum Likelihood-Constant False Alarm Rate) detector. That is, resorting to linear biparametric adaptive thresholds for both censoring and detection algorithms, we introduce a logarithmic amplifier to get a transformed Gaussian distribution. Assuming a homogeneous middle half ranked sub-SRW (Sliding Reference Window), we first compute the lower and upper censoring thresholds through the closed form solutions of the AML estimates of the unknown mean and standard deviation parameters. Then, upon censoring of both ends, we use the remaining set of data to estimate the unknown distribution parameters through the same expressions of the AML estimates to yield the detection threshold. Extensive simulations on both simulated SAR and real SAR images show that the AML-CFAR detector performs better than its competing state-of-the-art detectors.