Algorithm of steering kernel bilateral filtering

In order to keep edge character and avoid manual pre-definition of the parameters of existing bilateral filtering, a steering kernel bilateral filtering is proposed, which set the parameters of steering kernel adaptively. The steering kernel is designed based on the structure tensor and object scale. The direction angle and the proportion of axial length are set by structure tensor, and the length of short axis is obtained by calculating the object scale. Hence, the projection of the steering kernel is a large-scale circle in smooth regions; it is a slender oval in edge regions, and it changes to small-scale circle in corner regions. Simulation results show that the bilateral filtering based on the steering kernel can filter out the noise effectively and keep more edge characters. The algorithm is superior to traditional bilateral filter, gray-natural logarithm bilateral and the modified bilateral filters in both subjective and objective evaluations.