Flowing Bilateral Filter: Definition and Implementations

The bilateral filter plays a key role in image processing applications due to its intuitive parameterization and its high quality filter result, smoothing homogeneous regions while preserving the edges of the objects. Considering the image as a topological relief, seeing pixel intensities as peaks and valleys, we introduce a way to control the tonal weighting coefficients, the flowing bilateral filter, reducing “halo” artifacts typically produced by the regular bilateral filter around a large peak surrounded by two valleys of lower values. In this paper we propose to investigate exact and approximated versions of CPU and parallel GPU (Graphical Processing Unit) based implementations of the regular and flowing bilateral filter using the NVidia CUDA API. Fast implementations of these filters are important for the processing of large 3D volumes up to several GB acquired by x-ray or electron tomography.