Patch Confidence k-Nearest Neighbors Denoising

Cesario Vincenzo Angelino (1) , Eric Debreuve (1) , Michel Barlaud (1)

1 Laboratoire d'Informatique, Signaux, et Systèmes de Sophia-Antipolis (I3S) / Equipe IMAGES-CREATIVE

Recently, patch-based denoising techniques have proved to be very effective. Indeed, they account for the correlations that exist among patches of natural images. Taking a variational approach, we show that the gradient descent for the chosen entropy-based energy leads to a solution involving the mean-shift on patches. Then, we propose a patch-based denoising process accounting for the quality of denoising of each individual patch, characterized by a confidence. The denoised patches are combined together using each patch denoising confidence to form the denoised image. Experimental results show the better quality of denoised images w.r.t. NL means and BM3D. The proposed method has also been tested on a professional benchmark photography.

Mots clés: Denoising; image patch; confidence; entropy; mean-shift

Domaines: Traitement du signal et de l'image [eess.SP]; Traitement des images [eess.IV]; Traitement du signal et de l'image [eess.SP]

Cesario Vincenzo Angelino, Eric Debreuve, Michel Barlaud. Patch Confidence k-Nearest Neighbors Denoising. IEEE ICIP, Sep 2010, Hong Kong, Hong Kong SAR China. ⟨hal-00518078⟩