Maximum a posteriori probability background estimation
Thomas J. Green, William H. Payne, Vivian E. Titus, Eric J. Van Allen · The Journal of the Acoustical Society of America · 1999
Signal detection in acoustic data is complicated by a structured, nonuniform background that obscures signal characteristics. Consequently, efforts to estimate and whiten the background can yield significant improvements in downstream performance. In fact, the structure of the Neyman–Pearson optimal signal detector includes an explicit background whitening step. Linear estimators suffer from the difficulty of distinguishing between signal and background in the region of interest. Consequently, signal contamination can cause substantial errors in background estimates that reduce signal detectibility via subsequent algorithms and analysis. A nonlinear estimator has been developed that exploits models of both the measured data and the presumed scene and maximum a posteriori probability (MAP) theory to produce a background estimate. The data model is guided by the measurement physics and the scene model is based on a Markov random field assumption. Both models are augmented to accommodate anomalies. In this seminar the MAP estimator is described, along with practical issues associated with algorithm initialization, simulated performance with both white and colored-noise inputs, and real data examples. [Work supported by SPAWAR, under Air Force Contract F19628-95-C-0002.]