SAR Speckle Removal Based On Partial Differential Equations Method

Abstract: Synthetic Aperture Radar (SAR) images are contaminated by multiplicative noise because of its backscatters. Due to the coherence of the radar wavelength, speckle noise results in an severely important reduction in the efficiency of target detection and classification algorithms. Traditional noise-smoothing methods like Lee, Frost, Kuan etc. are not well suited to preserving edge structures in speckled images while reducing the speckle.Four conventional methods are introduced in chapter 2. Analyses and comparisons are listed completely. After deep study into the Partial Differential Equations (PDE's) and total variation principle, anisotropic & isotropic diffusion equations are casted into the frame of energy function diminishing. A new method of noise reduction then based on the partial differential equations (PDE) is derived from energy function diminishing frame in chapter 3.An improved method of speckle filtering in SAR image based on anisotropic diffusion equation with an edge indicator is presented in chapter 4. The edge indicator is used to force the pixel attributes to be the original value whenever it belongs to the region of features. The new diffusion function g() also plays an important role in the iteration scheme for numerical computations, which makes the original ill-posed problem regularized somehow. Theoretic and experimental results show that the new model can reduce the speckle in the homogeneous area inside the image without destroying important image features.In the last part of this thesis, we take a deeper look into the edge indicator. Toremove the isolated speckle noise and fake edge, a general approach for edge-preserving regularization using coupled PDE is obtained in chapter 5. This kind of approach can also be applied in general image processing. Comparison and experimental results show the proposed model's high performance.