Tese (doutorado)—Universidade de Brasília, Instituto de Ciências Exatas, Departamento de Ciência da Computação, 2015. / Submitted by Fernanda Percia França (fernandafranca@bce.unb.br) on 2016-02-23T15:22:45Z No. of bitstreams: 1 2015_EmersonLopesMachado.pdf: 3892504 bytes, checksum: 2b7bca95e649443f1f40cfc3df98fa56 (MD5) / Approved for entry into archive by Raquel Viana(raquelviana@bce.unb.br) on 2016-03-21T22:48:51Z (GMT) No. of bitstreams: 1 2015_EmersonLopesMachado.pdf: 3892504 bytes, checksum: 2b7bca95e649443f1f40cfc3df98fa56 (MD5) / Made available in DSpace on 2016-03-21T22:48:51Z (GMT). No. of bitstreams: 1 2015_EmersonLopesMachado.pdf: 3892504 bytes, checksum: 2b7bca95e649443f1f40cfc3df98fa56 (MD5) / Apresento nesta tese a análise teórica e as avaliações empíricas do conjunto de técnicas que proponho para a redução do custo computacional da classificação em tempo de teste de classificadores que são baseados em transformada e limiar suave aprendidos a partir dos dados de treinamento. Modificando o procedimento de otimização numérica utilizado na aprendizagem da transformada e do vetor de classificação, assim como aplicando um processamento em seus respectivos elementos após a aprendizagem, as técnicas que proponho permitem reduzir a quantidade de bits necessária para se realizar a classificação e trocar cada multiplicação em ponto flutuante por um simples deslocamento de bit em inteiro. Como estudo de caso, utilizei o algoritmo de classificação Learning Algorithm for Soft-Thresholding (LAST) e os mesmos conjuntos de dados utilizados no artigo que o apresenta. Os resultados do estudo de caso confirmam a possibilidade de se utilizar somente somas e deslocamentos em inteiro para a classificação em tempo de teste com uma perda de acurácia limitada. Essas operações de baixo custo computacional são importantes em implementações feitas em FPGA por permitir aumentar a velocidade de classificação e ao mesmo tempo diminuir o consumo de energia e o custo de fabricação. Além disso, as técnicas que apresento reduziram em quase 50% a quantidade de bits necessária para a extração de características na maioria dos experimentos que realizei. ______________________________________________________________________________________________ ABSTRACT / We present a theoretical analysis and empirical evaluations of a novel set of techniques for computational cost reduction of classifiers that are based on learned transform and soft-threshold. By modifying optimization procedures for dictionary and classifier training, as well as the resulting dictionary elements, our techniques allow to reduce the bit precision and to replace each floating-point multiplication by a single integer bit shift. We also show how the optimization algorithms in some dictionary training methods can be modified to penalize higher-energy dictionaries. We applied our techniques with the classifier Learning Algorithm for Soft-Thresholding, testing on the datasets used in its original paper. Our results indicate it is feasible to use solely sums and bit shifts of integers to classify at test time with a limited reduction of the classification accuracy. These low power operations are a valuable trade off in FPGA implementations as they increase the classification throughput while decrease both energy consumption and manufacturing cost. Moreover, our techniques reduced almost 50% of the bit precision in almost all datasets we tested.

This thesis focuses on image processing using wavelet transform. The usage of wavelet transform is analysed especially for image compression and image noise reduction purposes. The analysis describes in detail aspects and application of the following wavelet transform methods: CWT, DWT, DTWT, 2D DWT. The thesis further explains the meaning of the mother wavelet and studies certain specific kinds of wavelets, kinds of thresholding and its purposes and also touches on the JPEG2000 standard. Mathematica and Sage software packages were used to design algorithms for image compression and image noise reduction, utilising relevant wavelet transform findings. The concluding part of the thesis compares the two software packages and results obtained using different algorithms.

This thesis focuses on image processing using wavelet transform. The usage of wavelet transform is analysed especially for image compression and image noise reduction purposes. The analysis describes in detail aspects and application of the following wavelet transform methods: CWT, DWT, DTWT, 2D DWT. The thesis further explains the meaning of the mother wavelet and studies certain specific kinds of wavelets, kinds of thresholding and its purposes and also touches on the JPEG2000 standard. Mathematica and Sage software packages were used to design algorithms for image compression and image noise reduction, utilising relevant wavelet transform findings. The concluding part of the thesis compares the two software packages and results obtained using different algorithms.