An image driven bilateral filter with adaptive range and spatial parameters for denoising Magnetic Resonance Images

Bilateral filter is an edge preserving denoising filter, widely used in medical image processing and raster graphics editors. Its performance greatly depends upon selection of spatial and radiometric parameters. Subjective selection of these parameters through trial and error can be sub-optimal. In this paper, an image driven method to compute the optimum values of spatial and radiometric parameters is proposed. Local optima of the parameters are adaptively estimated from the local grey level dispersion. The proposed method exhibited a Noise Suppression Ratio (NSR), Edge preservation Index (EPI) and Edge Strength Similarity Index Metric (ESSIM) of 0.62 ± 0.15, 0.76 ± 0.04 and 0.9998 ± 0.00005, respectively. It is superior to the method of computing the operational parameters from the standard deviation of noise in which the values of the performance matrices are NSR = 0.556 ± 0.223, EPI = 0.72 ± 0.01 and ESSIM = 0.9998 ± 0.0001. The bilateral filter with locally adaptive spatial and radiometric parameters could selectively denoise homogeneous regions, without degrading the morphological edges.