Filtragem não-local do ruído Speckle em imagens SAR com modelagem estatística dos coeficientes da Wavelet de Haar e distâncias estocásticas

• Autor(a) principal: Penna, Pedro Augusto de Alagão
• Data de Publicação: 2018
• Tipo de documento: Tese
• Idioma: por
• Título da fonte: Repositório Institucional da UFSCAR
• Texto Completo: https://repositorio.ufscar.br/handle/ufscar/10195
• topic: Distâncias estocásticas; Distribuição EP; Filtragem; SAR; NLM; Speckle; Despeckling; Wavelets; CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO::METODOLOGIA E TECNICAS DA COMPUTACAO

Due to the coherent processing of synthetic aperture radar (SAR) systems, multiplicative speckle noise arises providing a granular appearance in SAR images. This kind of noise makes it difficult to analyse and interpret Earth surface images. Therefore, the search for new techniques to mitigate the speckle is a constant task in the image processing literature. Current state-of-the-art filters in remote sensing area explore the philosophy of similarity between patches (neighborhoods). This thesis aims to expand a recently proposed filtering algorithm:the Non-Local Means (NLM), which represents a new paradigm for filtering images, and analyses and compares its capacity of speckle reduction in intensity SAR images, technique known in the literature as despeckling. This filter was originally proposed for the additive white Gaussian noise (AWGN). The NLM filter extension considers a scenario with the more aggressive noise, i.e., the single-look speckle, and it is possible to attenuate the noise by replacing the original distance used to measure the similarity between patches, the Euclidean distance, with the stochastic distances and apply the proposed filter in the Haar wavelets domain. To achieve this goal, the Haar wavelet coefficients were described by the Exponential-Polynomial (EP) and Gamma distributions. The main contribution of this proposal is to work with the NLM, originally developed for the image space domain, directly in the wavelets domain, by computing the stochastic distances based on the EP and Gamma distributions. In addition, this proposal shows that it is advantageous to use elaborate methods for post-filtering processing, such as Dual Domain Filtering (DDF) and Data Adaptive Dual Domain Denoising (DA3D), as they can improve the quality of the filtered image both of the proposed method and the state-of-the-art algorithm. Finally, an analysis and comparison of the results of the proposed method are made, which show that this new approach was able to attenuate the presence of speckle in the more aggressive case, with some of the recent filters in the literature. The obtained results show that the proposed filter is competitive.