SAR speckle noise removal and image compression based on anisotropic diffusion and wavelets

Unlike Additive White Gaussian Noise (AWGN), speckle noise in Synthetic Aperture Radar (SAR) images is multiplicative in nature and has an asymmetric probability distribution function (pdf). The behavior of speckle noise in wavelet do mains is very different with AWGN. Using the Haar basis as an example, a complete statistical analysis of speckle noise in wavelet domains is shown in its distribution. By exploring the properties of Discrete Wavelet Transform (DWT) and Anisotropic Diffusion (AD), several novel algorithms have been developed to remove the speckle noise in SAR images. It is well-known that wavelets are sensitive to the high-frequency components in an image, including edges, textures and noise. In the Anisotropic Diffusion process, we expect efficient diffusion occurring at pixels of speckle noise, while, “inverse” diffusion is expected to occur at pixels of textures and edges. The Anisotropic Diffusion process is adaptively controlled by edges, textures and noise information in wavelet domains. This approach is very efficient for SAR, images with heavy speckle noise. Spatial Oriented Tree (SOT) structure is a well-known structure for wavelet-based image compression. We have developed an SOT-based technique to remove multiplicity of speckle noise. Based on this technique, new algorithm combining AD and the SOT-based technique has been developed for the speckle noise removal in SAR images. The emerging Bayesian wavelet estimation techniques are also adapted for the speckle noise removal in SAR images. This dissertation also reports a novel algorithm that combines wavelet-based despeckled technique and Set Partitioning In Hierarchical Trees (SPIHT) algorithm for simultaneous noise removal and image compression for SAR images.