Aplicação de computação estocástica à detecção de bordas em imagens digitais

• Ano de defesa: 2019
• Autor(a) principal: Cavalcanti, Danilo Barreto
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal da Paraíba; Brasil; Engenharia Elétrica; Programa de Pós-Graduação em Engenharia Elétrica; UFPB
• Palavras-chave em Português: Computação estocástica; Processamento de imagens; Detecção de bordas; Eficiência energética; Stochastic computing; Image processing; Edge detection; Energy efficiency; CNPQ::ENGENHARIAS::ENGENHARIA ELETRICA
• Link de acesso: https://repositorio.ufpb.br/jspui/handle/123456789/19380

This work compares the grey image edge detection algorithm based on Sobel operators considering two implementations: conventional digital logic and Stochastic Computing. The latter refers to an unconventional way of organising and operating numbers, treating them as probabilities. There are advantages in noise robustness, integrated circuit area utilisation and power consumption, when compared to CMOS Digital Electronics, in certain applications. This work implemented and evaluated filters in software (Python) and hardware (Verilog/FPGA). It is shown that the order of magnitude of the difference between results in both versions is of two bits. It is shown that the stochastic approach is more robust to soft errors, while also having less discrepant error distributions between 1% and 5% of error bit rates. Finally, the number of logical gates and power required for each implementation is presented. This stablishes guidelines of cost-benefit for energy efficient Digital Electronics projects.