Une approche automatisée pour l'élimination du bruit quasi-périodique dans les images naturelles
Frédéric Sur, Michel Grédiac · HAL (Le Centre pour la Communication Scientifique Directe) · 2015
Quasi-periodic noise may affect digital images. This phenomenon is reflected by spurious repetitive patterns covering the whole image. Quasi-periodic noise is by nature well localized in the Fourier spectrum. A possibility is thus to attenuate it through a well-designed notch filter. In contrast to existing algorithms which require hand-tuned filter design, this research report presents an automated approach based on the expected power spectrum of a natural image. The resulting algorithm enables to eliminate a large range of repetitive structures, from simple periodic noises, whose influence on the image spectrum is limited to a few Fourier coefficients, to quasi-periodic noises with much more complex spectrum structures. The proposed algorithm is assessed on the basis of various experiments. A comparison with morphological component analysis (MCA), a blind source separation algorithm, is also provided. A Matlab implementation is available. This report is anextension of a published article.