Filtragem de ruído Poisson utilizando o algoritmo BM3D com distâncias estocásticas e transformações de estabilização de variância

• Autor(a) principal: Abegão, Rodrigo Garcia
• Data de Publicação: 2024
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Repositório Institucional da UFSCAR
• Texto Completo: https://hdl.handle.net/20.500.14289/22232
• topic: Poisson noise; Stochastic distances; CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO::METODOLOGIA E TECNICAS DA COMPUTACAO; BM3D; VST; Ruído poisson; Distâncias estocásticas

Noise in images is a phenomenon present in several fields of academic research, dueto the fact that its presence in them can impair the analysis and interpretation of datacontained in them. Thus, generating significant implications in areas such as: medicine,biological sciences, astronomy and geology, among others.In the medical field, images generated by tomographs are degraded by a noise calledPoisson. This is a statistical and signal-dependent noise caused by the absorption andscattering phenomena caused during the interaction between the X-ray beam and matter.This type of noise is caused by several sources, including variations in light intensity,measurement errors, electromagnetic interference and electronic noise.There are several filters to mitigate the presence of Poisson noise in images. And, themost used techniques in this environment are non-local, such as BM3D. The purpose ofwhich is the similarity between small equivalent fragments in the image, called patchessuch that the state of the art generally adopts the Euclidean distance in cases of AWGN.This research proposes the use of stochastic distances in the BM3D algorithm toachieve results superior to the current state of the art. The incorporation of stochasticdistances enables the exploration of innovative approaches that take into account differentcharacteristics of data distributions. This approach significantly expands the possibilitiesfor analyzing and comparing random variables, providing a more comprehensive and detailedunderstanding of the relationships and patterns present in the data. To achievethese objectives, the variance-stabilizing technique known as VST was used to transformPoisson noise into AWGN. Several studies demonstrate the effectiveness of using differentdistances, although performance depends on the specific type of noise present in the data.