Content adaptive bilateral filtering

Bilateral filter is perhaps the most popular and simplest edge-preserving local filter. However, it may produce halos near some edges due to unwanted smoothing of the edges. In this paper, a content adaptive bilateral filter is proposed to overcome this problem. Both spatial similarity and range similarity parameters of the proposed bilateral filter are adaptive to the content of an image to be filtered rather than being fixed as in the existing bilateral filter. The proposed filter preserves edges and smoothes flat areas better than the existing bilateral filter. It is applied to design a local tone mapping algorithm for high dynamic range images as well as a noise reduction algorithm for low dynamic range color images. Experimental results show that the resultant tone mapping algorithm produces images with better visual quality and at the same time halo artifacts are avoided from appearing in the tone mapped image. The resultant noise reduction algorithm can also produce images with higher structural similarity index values and better visual quality. Noise in smooth regions are more reduced and edges are preserved better in the de-noised image.