HIGH-RESOLUTION ALGORITHMS FOR LOCATING CLOSELY SPACED OBJECTS VIA INFRARED FOCAL-PLANE ARRAYS

The location of a single point source in infrared imaging is typically achieved through conventional methods such as centroiding. More chall enging problems with multiple point sources require alternative locati on-finding methods with the potential of resolving closely spaced obje cts. The authors introduce an algorithm predicated on least-squared-er ror (LSE) modeling with a Gram-Schmidt orthogonalization step. Its noi se performance is compared with two other high-resolution algorithms b ased on the eigendecomposition of the input data. Estimates obtained t hrough the LSE modeling approached the Cramer-Rao lower bound for high signal-to-noise ratios. However, its performance is severely degraded in the presence of non-Gaussian noise. An outlier detection scheme th at may be used in conjunction with the location and amplitude estimati on procedure is described. Its effectiveness is demonstrated through M onte Carlo simulations.