Modificação no termo de regularização realçando os elementos de borda do modelo autorregressivo simultâneo de super-resolução multiframe

• Ano de defesa: 2019
• Autor(a) principal: Campana, Vitor Faical
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal do Espírito Santo; BR; Doutorado em Engenharia Elétrica; Centro Tecnológico; UFES; Programa de Pós-Graduação em Engenharia Elétrica
• Palavras-chave em Português: Processamento de imagens; Super-resolução; Multiframe; Inferênia Bayesiana variacional; SAR; Descritor fractal estatístico; Regularização; Image processing; Super-resolution; Variational Bayesian inference; Statistical fractal descriptor; Regularization; subject.br-rjbn; Engenharia Elétrica
• Link de acesso: http://repositorio.ufes.br/handle/10/13664

There are several methods of super-resolution in the literature, based on increasing the resolution of the image from a single low resolution image (single-frame) or from a set of low resolution image (multiframe). This work studies a method of multiframe super-resolution, especifically the method called simultaneous autoregressive (SAR), which is based on the variational Bayesian inference approach. Following this study, a modification of this method is proposed, which includes a high-order statistical fractal descriptor in its formulation. This modification causes a change in the a priori distribution model of the SAR method. More precisely, it is proposed to use the descriptor Local Morphologic Multifractal Exponent (LMME) to estimate the hyperparameters of the probability distribution function which represents the high resolution (HR) image. Due to the texture sensitivity feature of the LMME, the estimated HR images showed good improvements in their edge elements and detail regions when compared to images generated by the original SAR model, while mitigating unwanted noise. The proposed modification was also extended to the model combining the SAR method with the ℓ1–norm. A comparison using peak-signal-to-noise ratio (PSNR) and structural similarity index (SSIM) metrics is presented to measure the quality of the HR images estimated by the proposed methods in relation to others presented in the literature, within the multiframe approach. As a result, the use of the LMME descriptor presented better adjustments of the parameters of the estimation distributions of the restored images, reflected in the values of PSNR and SSIM obtained in the experiments using low resolution images with moderate levels of noise (SNR of 30 dB and 40 dB). In addition, the use of the LMME descriptor in the SAR model made the Expectation and Maximization (EM) algorithm converge in a smaller number of iterations.