Bilateral filter

The bilateral filter is defined as : I^\text{filtered}(x) = \frac{1}{W_p} \sum_{x_i \in \Omega} I(x_i)f_r(\|I(x_i) - I(x)\|)g_s(\|x_i - x\|), and normalization term, {W_p}, is defined as : W_p = \sum_{x_i \in \Omega}{f_r(\|I(x_i) - I(x)\|)g_s(\|x_i - x\|)} where : I^\text{filtered} is the filtered image; : I is the original input image to be filtered; : x are the coordinates of the current pixel to be filtered; : \Omega is the window centered in x, so x_i \in \Omega is another pixel; : f_r is the range kernel for smoothing differences in intensities (this function can be a Gaussian function); : g_s is the spatial (or domain) kernel for smoothing differences in coordinates (this function can be a Gaussian function). The weight W_p is assigned using the spatial closeness (using the spatial kernel g_s) and the intensity difference (using the range kernel f_r). Consider a pixel located at (i, j) that needs to be denoised in image using its neighbouring pixels and one of its neighbouring pixels is located at (k, l). Then, assuming the range and spatial kernels to be Gaussian kernels, the weight assigned for pixel (k, l) to denoise the pixel (i, j) is given by : w(i, j, k, l) = \exp\left(-\frac{(i - k)^2 + (j - l)^2}{2 \sigma_d^2} - \frac{\|I(i, j) - I(k, l)\|^2}{2 \sigma_r^2}\right), where σd and σr are smoothing parameters, and I(i, j) and I(k, l) are the intensity of pixels (i, j) and (k, l) respectively. After calculating the weights, normalize them: : I_D(i, j) = \frac{\sum_{k, l} I(k, l) w(i, j, k, l)}{\sum_{k, l} w(i, j, k, l)}, where I_D is the denoised intensity of pixel (i, j). == Parameters ==

• As the range parameter σr increases, the bilateral filter becomes gradually equivalent to Gaussian blur; i.e. σd becomes dominant, because the gaussian kernel is then mostly flat. • As the spatial parameter σd increases, the larger features get smoothened. == Limitations ==

The bilateral filter in its direct form can introduce several types of image artifacts: • Staircase effect – intensity plateaus that lead to images appearing like cartoons • Gradient reversal – introduction of false edges in the image. There exist several extensions to the filter that deal with these artifacts, like the scaled bilateral filter that uses downscaled image for computing the weights. Alternative filters, like the guided filter, have also been proposed as an efficient alternative without these limitations. == Implementations ==

Adobe Photoshop implements a bilateral filter in its surface blur tool. GIMP implements a bilateral filter in its Filters → Blur tools; and it is called Selective Gaussian Blur. The free G'MIC plugin Repair → Smooth [bilateral] for GIMP adds more control. A simple trick to efficiently implement a bilateral filter is to exploit Poisson-disk subsampling. OpenCV implements the function: bilateralFilter( source, destination, \Omega, f_r, g_s ). ==Related models==

The bilateral filter has been shown to be an application of the short time kernel of the Beltrami flow that was introduced as an edge preserving selective smoothing mechanism before the bilateral filter. Other edge-preserving smoothing filters include: anisotropic diffusion, weighted least squares, edge-avoiding wavelets, geodesic editing, guided filtering, and domain transforms. == See also ==