Proposição e avaliação de indicadores de desempenho para algoritmos de restauração de imagens / Proposition and evaluation of different performance indicators for image processing algorithms

AUTOR(ES)
DATA DE PUBLICAÇÃO

2006

RESUMO

The restoration of images acquired with Atomic Force Microscopes (AFM) consists on a very intense research area in nanotechnology. During the acquisition process the AFM images are subjected to the constraints imposed by the experimental apparatus usually presenting poor signal to noise ratios as well as the effects of tip-sample interaction. In order to process images related to biological structures, several restoration algorithms have been proposed and tested, and in the present dissertation is used a method based in Tikhonovs regularization. Throughout the development of several restoration algorithms it has been observed that the usual performance measures did not correspond to the human visual perception of the restored images. In the present dissertation are proposed and evaluated several performance measures for restoration algorithms used in the post-processing of images acquired in nanoscale with AFM as well as artificial images created in order to test the performance measure used. Usually used performance measures are investigated as well as measures based on Bregman distances. The research was conducted in two parts. First a set of images was generated and restored using Tikhonovs regularization functional, being then distributed to 50 human referees. Their task consisted in evaluating the images in order to grade how close the restored images were to the original images, in respect to human visual perception. The second part of the research consisted in the implementation of a computational program to perform the computation of the mathematical performance measures. Afterwards a comparison was performed of the mathematical performance measures with respect to the visual perception of the human referees, being the latter essentially qualitative which was then transformed in a numerical scale in order to allow the comparison with the mathematical measures. It was observed that depending on the parameters used in the restoration algorithm based in Tikhonovs regularization functional and the type of the restored image, the mathematical performance measures based on Bregman distances present better results than those obtained with other usual performance measures such as the Mean Square Error (MSE) yielding to a better agreement with the evaluation performed by the human referees.

ASSUNTO(S)

image processing - algorithms tikhonovs regularization human referees indicadores matemáticos avaliadores humanos microscopia de força atômica performance measures matematica aplicada processamento de imagens - algoritmos atomic force microscopy regularização de tikhonov restauração de imagens image restoration

ACESSO AO ARTIGO

Documentos Relacionados