Banca de TCC: Vinícius Rodrigues dos Santos

UNIVERSIDADE FEDERAL DE PELOTAS
CENTRO DE DESENVOLVIMENTO TECNOLÓGICO
TRABALHO DE CONCLUSÃO DE CURSO

Apresentações Finais (2016/1)

Int-DWTs Library – Algebraic Simplifications Increasing Performance and Exactitude of Discrete Wavelet Transforms
por
Vinícius Rodrigues dos Santos

Curso:
Ciência da Computação

Banca:
Profa. Renata Hax Sander Reiser (orientador)
Prof. Maurício Lima Pilla (co-orientador)
Prof. Alice de Jesus Kozakevicius (co-orientador)
Prof. Adenauer Correa Yamin
Profa. Aline Brum Loreto
Prof. Marilton Sanchotene de Aguiar

Data: 11 de Julho de 2016

Hora: 10:00h

Local: Pós 2, FAT

Resumo do Trabalho:

This project describes the main results consolidating the Interval Methods of the Discrete Wavelet Transforms Library (Int-DWTs), providing algebraic simplifications in order to increase performance and guarantee exactitude of Discrete Wavelet Transforms (DWTs). The methods in the Int-DWTs include interval extensions and corresponding optimizations to the Haar Wavelet Transform (HWT) and Daubechies Wavelet Transform (DWT), which are implemented by using C-XSC. Both steps in the HWT, meaning as Cascade and à trous algorithms, are extended to the interval approach. Optimizations for the normalized formulations of the one- and two-dimensional transforms to increase speed and guarantee accuracy of results are also presented. As an application, image compression is addressed, whose quality is computed and compared by different metrics. The error analysis of the different interval formulations as well as the speedup obtained through the interval formulations and the proposed simplifications are analyzed. The results consider a study over the interval extension of DWT, deducting simplifications and implementing optimizations based on the analogous methodology performed over HWT approaches. In addition, not only interval algorithms are developed but also metrics for evaluating procedure’s accuracy are considered. By assuming concepts from Interval Mathematics, it enables us to perform interval analysis and efficiently manage of computing errors of DWTs. The metrics being used to measure result quality in the Int-DWTs are the following: Euclidean Distance, Mean Squared Error and Peak signal-to-noise ratio. In contexts that the scales of representation are relevant to the problem, the accuracy gain obtained with the proposed optimizations in the Int-DWTs represent a significant contribution to the scientific computing research area.

Para mais informações acesse: http://inf.ufpel.edu.br/notcc/doku.php?id=bancas:2016_1