Patch-based bilateral filter: Algorithms and performance

The bilateral filter has a weakness in its denoising ability, because the intensity similarity is evaluated by comparing just the intensity values of a given pixel and another pixel. To alleviate this problem, this paper proposes a patch-based bilateral (PBL) filter in which the intensity similarity is evaluated by comparing a whole patch around each pixel. Experimental results for real images show that this method outperforms the bilateral filter in all cases, is comparable to the nonlocal-(NL-)means filter, and outperforms the more elaborate variants of the NL-means filter in some cases.