Fast bilateral filter with spatial subsampling

Yang Yang, Yiwen Xiong, Yanqing Cao, Lanling Zeng, Yan Zhao, Yongzhao Zhan

Abstract: The bilateral filter is a non-linear edge-preserving filter that can be adopted in a variety of tasks in computer photography. However, the naive bilateral filter is computationally expensive. Existing researches on the acceleration of bilateral filter mostly concentrate on range approximation. Nevertheless, the range kernel has more impact on the bilateral filter than the spatial kernel. Range approximation would have more side effects. In this paper, we propose a novel approximation of the bilateral filter with spatial subsampling, where the affinity matrix is estimated from a subset of it. We show that the main computational burden of our approximation is a large linear system, for which we propose an efficient iterative algorithm to solve. We have carried out both quantitative and qualitative experiments to evaluate our fast bilateral filter. Experimental results suggest that the proposed filter outperforms the state-of-the-art methods in approximation accuracy. The proposed filter is highly efficient; under a moderate sampling rate, i.e., \((1/5)\times (1/5)\), it needs 0.29s to process a color image with 1 megapixel on an Intel i7-9700 CPU.

External IDs:dblp:journals/mms/YangXCZZZ23