Development of algorithm for optimally weighted non-iterative bilateral filter (OW-NI-BF) to eliminate gaussian noise

Abstract

Gaussian filter is linear that smooth independent of edges and details of the image. The standard bilateral filter is effective for low-density noise. However, the performance of filter degrades with an increase in noise level. We designed a non-iterative bilateral filter algorithm which is robust for large noise levels, but lacks low noise levels the performance. To get the best denoising performance out of these two, standard bilateral filter and proposed non-iterative bilateral filter is combined in weighted fashion using Stein's unbiased risk estimate. Thus, the proposed optimally weighted non-iterative bilateral filter (OW-NI-BF) algorithm is guaranteed to perform better than either of the component filter. It is non-linear, local and non-iterative that works on geometric closeness and gray level similarities irrespective of smoothing edge. In Gaussian noise scenarios, the performance of the proposed OW-NI-BF algorithm is compared with various methods. Quantitative and visual denoising results demonstrate significant improvement over the original filter. PSNR and IQI are used to measure the quality of the de-noised image.