Synthetic Aperature LADAR: A Model-Based Approach

Synthetic aperture LADAR (SAL) allows high resolution imaging of distant objects. Basic SAL image processing is based on fast Fourier transform (FFT) techniques originally developed for use in radar. These techniques can amplify noise and limit resolution. More advanced reconstruction techniques have been proposed for synthetic aperture radar (SAR), but have not been adapted for SAL. In addition, both conventional SAL and advanced SAR algorithms reconstruct the complex-valued reflection coefficient instead of the real-valued reflectance which leads to speckled images. In this paper, we present a model-based iterative reconstruction (MBIR) algorithm designed specifically for SAL. Rather than estimating the reflection coefficient, we propose estimating its variance which is equal to the reflectance, a function that more closely resembles conventional optical images. A Bayesian framework is used to find the maximum a posteriori (MAP) estimate for the reflectance using a Q-Generalized Gaussian Markov random field (QGGMRF) prior model. The QGGMRF is able to model complex correlations between neighboring pixels which promotes a smooth and more natural looking image. The expectation-maximization (EM) algorithm is used to derive a surrogate for the MAP cost function. Finally, the proposed MBIR algorithm is tested on both simulated and experimental data. Results show significant and consistent improvements over existing reconstruction techniques in terms of image contrast, speckle reduction, autofocus, and low signal-to-noise ratio performance.