Anais do X SIBGRAPI (1997) #ART74


Um Método Baseado na Taxa de Falsas Descobertas de Coeficientes para a Redução de Ruído em Imagens no Domínio da Transformada de Wavelets

Weneidner Moura de Souza, Universidade de Brasília - Depto. de Eng. Elétrica
Daniel Távora de Queiroz Cobra, Gyron Sistemas Autônomos Ltda.

Abstract. We propose a technique for the reduction of speckle noise in images based on the method of False Discovery Rate of Coefficients (FDRC). This is a de-noising algorithm, i. e., an algorithm for noise reduction in the wavelet transform domain. We have extended the FDRC algorithm to the two-dimensional case and have incorporated the cycle-spinning technique in order to deal with the pseudo-Gibbs phenomenon. Our technique can thus be classified as a second-generation de-noising algorithm. The method is adaptive in nature and is based on hypothesis testing, seeking to control the expected proportion of wavelet coefficients that are incorrectly included among those used to reconstruct the image. We provide an example of the application of the proposed method to the reduction of speckle noise in side-scan sonar images. The results show that the method is able to significantly reduce speckle noise while preserving the details of the image.

[ full version | contact authors ]