JEDI: Adaptive Stochastic Estimation for Joint Enhancement and Despeckling of Images for SAR Wen Zhang wen.zhang@ieee.org Alexander Wong alexanderwong@ieee.org David A. Clausi dclausi@uwaterloo.ca Vision and Image Processing Group Systems Design Engineering University of Waterloo Waterloo, Ontario, Canada N2L 3G1 Abstract Synthetic aperture radar (SAR) images are degraded by a form of multiplicative noise known as speckle. Current methods for despeckling are limited in that they either do not perform enough noise attenuation, or do not adequately preserve or enhance image detail. We propose a novel adaptive stochastic method for joint enhancement and de- specking of images (JEDI) for SAR. The proposed method utilizes an adaptive importance sampling scheme based on local statistics to generate random samples while reducing estimation variance. A Monte Carlo estimate is computed based on the generated samples, wherein the samples are aggregated to form a despeckled and detail-enhanced re- sult. The advantage of JEDI is the ability to efficiently take advantage of information redundancy in speckled images to reduce the effects of speckle while simultaneously en- hancing detail visualization. Testing with both simulated and real speckled images shows that JEDI typically out- performs popular despeckling algorithms such as Frost fil- tering, anisotropic diffusion, median filtering, Γ-MAP and GenLik in terms of quantitative and qualitative visual qual- ity. On average, JEDI provides a 2-15% improvement in PSNR and a 5-14% improvement in image quality index measures over the tested methods. 1 Introduction Synthetic aperture radar (SAR) is a widely-used tech- nology in remote sensing applications. SAR images are generated by measuring the backscattered signal from a ra- dio pulse. Because of phase coherence effects caused by multiple scatterers in a resolution cell, the resulting images are characterized by a grainy pattern known as speckle [1]. Popular speckle reduction methods include linear least- squares estimators (LLSE) based on local statistics, such as the Lee [2], Frost [3], and Kuan [4] filters. Other despeck- ling methods include Γ-MAP [5], anisotropic diffusion [6], adaptive weighted median filtering [7], and wavelet-domain filtering (GenLik) [8]. A limitation of current despeckling methods is insuffi- cient noise attenuation in homogeneous regions, especially with correlated speckle. Moreover, while current despeck- ling methods are designed to preserve edges, they do not enhance them for better visibility, and often do not provide speckle reduction in edge regions. These problems can limit the visual quality of the despeckled image, as well as create difficulties for automatic segmentation techniques, to such an extent that SAR smoothing is not performed in these ap- proaches [9]. The current methods all depend, to varying extents, on local information redundancy. Recently, Wong et al. [10] proposed a Monte Carlo estimation framework for denois- ing that allows efficient use of global information redun- dancy. Based on this, we propose a novel method for joint enhancement and despeckling of images (JEDI) that em- ploys an adaptive stochastic approach to overcome limita- tions inherent in local methods. The proposed JEDI method is able to attain greater levels of speckle attenuation while increasing the visibility of image structures. 2 Adaptive Stochastic Estimation 2.1 Speckle Model Let S be the discrete lattice on which the images f , g, and random field n are defined, and let x be an index into the lattice. Speckle, which arises from the constructive and destructive interference of the backscattered signal, can be modeled as multiplicative noise [1] according to g(x)= f (x) · n(x), (1) 2009 Canadian Conference on Computer and Robot Vision 978-0-7695-3651-4/09 $25.00 © 2009 IEEE DOI 10.1109/CRV.2009.14 101