Autor(es):
Francisco Paulo Marques de Oliveira 
; Todd C. Pataky ; João Manuel Ribeiro da Silva Tavares
Data: 2010
Identificador Persistente: http://hdl.handle.net/10216/43857
Origem: Repositório Aberto da Universidade do Porto
Assunto(s): Ciências Físicas; Física; Biofísica; Biomecânica
Descrição
In this study, we compare the performance of six fully automatic methods of within-subjects pedobarographic image registration: principal axes, modal matching, min(XOR), min(MSE), contours-based and frequency-based. These algorithms were tested on 30 control image pairs considered in previous studies. The accuracy was assessed by visual inspection and using the image similarity measures: exclusive-or (XOR) and mean squared error (MSE). Visually, we did not find differences in the registration accuracy among the min(XOR), min(MSE), contours-based and frequency-based algorithms. On the other hand, using the similarity measures, we found out that the best XOR value was achieved by the contours-based algorithm, closely followed by the min(XOR). Additionally, the best MSE value was achieved by the min(MSE) algorithm, nearly followed by the frequency-based algorithm. Finally, the algorithms based on principal axes and modal matching revealed low robustness.
