Implementação paralela do algoritmo iterativo de busca do parâmetro de regularização ótimo para o funcional de Tikhonov no problema de restauração de imagens

• Autor(a) principal: Oliveira, Claudir
• Data de Publicação: 2012
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Biblioteca Digital de Teses e Dissertações da UERJ
• Texto Completo: http://www.bdtd.uerj.br/handle/1/13761
• topic: Parameter of regularization; Iterative method of search; Otimização; Restoration of images; Parâmetro de regularização; Método iterativo de busca; Restauração de imagens; Computação Paralela; Processamento de imagens Técnicas digitais; Algoritmos; Realimentação (Eletrônica); CNPQ::CIENCIAS EXATAS E DA TERRA::MATEMATICA::MATEMATICA APLICADA

The use of techniques with the functional of Tikhonov in image processing has been widely used in recent years. The basic idea in this process is to modify an initial image using a convolution equation and to find a parameter which minimizes the function in order to obtain an aproximation of the original image. However, a typical problem in this method consists in the choice of the regularization parameter in the appropriate compromise between accuracy and stability of the solution. A method developed by researchers of IPRJ and UFRJ, operating in the area of inverse problems, consists on minimizing a functional of residues through a functional parameter Tikhonov s regularization. A strategy that uses the iterative search of this parameter aiming at to get a minimum value for the functional in the following iteration was adopted recently in a serial algorithm of restoration. However, the computational cost is a factor problem found when using the iterative search. With this approach, an implementation in C++ language was made using techniques of parallel computation using MPI (Message Passing Interface) for the in the choice of the regularization parameter in the appropriate compromise between accuracy and stability of the soluton with the method of iterative search, thus reducing, the time of execution required for the algorithm. A modified version of the Jacobi method is considered to be two versions of the algorithm, one serial and one parallel. This algorithm is adequate for parallel implementation because it has no data dependencies such as the Gauss-Seidel method is also shown to converge. As indicating of performance for evaluation of the restoration algorithm, in addition to the traditional measures, new metric that is based on subjective criteria called IWMSE (InformationWeighted Mean Square Error) is used. These metrics were introduced in the program of image processing and allow to make the analysis of the restoration to each step of iteration. The results obtained using the two possible versions verify the efficiency of acceleration and the parallel implementation. The method of parallelism achieved satisfactory results in a shorter processing time and with acceptable performance.