The random noise arising from the acquisition process of magnetic resonance images negatively influences the diagnostic process. Thus, a denoising step is commonly adopted in case of many image processing and analysis tasks. The main criticism of denoising filters is to preserve edges and details while reducing noise across the image. Within this manuscript the a novel denoising filter, namely the KS-NLM, is proposed, with the aim of improving the denoising quality. The approach belongs to the Non Local Means (NLM) family as it exploits groups of similar pixels. The novelty consists in the similarity measurement, that is based on the statistical distribution of data instead of the similarity between the local textures of the image. In particular, the Cumulative Distribution Function for each pixel is evaluated from the acquired data, and subsequently the Kolmogorov-Smirnov distance is computed as similarity metric. Similar pixels are subsequently fused. The method has been tested on a real dataset acquired via a 3T scanner, and its performances have been compared to other state-of-art filters.
Magnetic Resonance Imaging Restoration based on Kolmogorov-Smirnov Non Local Mean
BASELICE, FABIO;FERRAIOLI, GIAMPAOLO;SORRISO, ANTONIETTA
2017-01-01
Abstract
The random noise arising from the acquisition process of magnetic resonance images negatively influences the diagnostic process. Thus, a denoising step is commonly adopted in case of many image processing and analysis tasks. The main criticism of denoising filters is to preserve edges and details while reducing noise across the image. Within this manuscript the a novel denoising filter, namely the KS-NLM, is proposed, with the aim of improving the denoising quality. The approach belongs to the Non Local Means (NLM) family as it exploits groups of similar pixels. The novelty consists in the similarity measurement, that is based on the statistical distribution of data instead of the similarity between the local textures of the image. In particular, the Cumulative Distribution Function for each pixel is evaluated from the acquired data, and subsequently the Kolmogorov-Smirnov distance is computed as similarity metric. Similar pixels are subsequently fused. The method has been tested on a real dataset acquired via a 3T scanner, and its performances have been compared to other state-of-art filters.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.