A Non-Local Algorithm for Image Denoising

Patch-based segmentation using expert priors: Application to hippocampus and ventricle segmentation

“…In the proposed method, the intensity-based distances between the patch under study and the patches in the training subjects are used to perform a weighted label fusion based on the nonlocal means estimator (Buades et al, 2005). The term nonlocal indicates that the spatial distance between the patches' centers is not taken into account; thus, the weight of each sample is only driven by the similarity of intensities between patches.…”

“…• Search volume definition Initially, the nonlocal means denoising filter was proposed as a weighted average of all the pixels in the image (Buades et al, 2005). For computational reasons, the entire image cannot be used and the number of pixels involved has to be reduced.…”

“…For computational reasons, the entire image cannot be used and the number of pixels involved has to be reduced. As done for denoising (Buades et al, 2005;Coupe et al, 2008), we use a limited search volume V i , defined as a cube centered on the voxel x i under study. Thus, within each of the N selected subjects, we search for similar patches in a cubic region around the location under study (see Fig.…”

“…As exemplars, patch-based methods are currently the focus of attention of the computer vision community in various domains such as texture synthesis (Efros and Freeman, 2001), in-painting (Criminisi et al, 2004), restoration (Buades et al, 2005), and single-frame super resolution (Protter et al, 2009). In each of these domains, patch-based methods have been the subject of intensive investigation because they exhibit very high performance despite their simplicity.…”

“…In each of these domains, patch-based methods have been the subject of intensive investigation because they exhibit very high performance despite their simplicity. Inspired by the nonlocal means denoising filter (Buades et al, 2005), we propose a nonlocal patch-based approach using expert manual segmentations as priors in the context of anatomical segmentation. The nonlocal means filter has two interesting properties that can be exploited to improve segmentation.…”