Pyramid Texture Filtering
Qing Zhang1Hao Jiang1Yongwei Nie2Wei-Shi Zheng*1
1Sun Yat-sen University,  2South China University of Technology 

ACM Transactions on Graphics (Proceedings of SIGGRAPH 2023)

We demonstrate texture filtering (also referred to as structure-preserving filtering) based on Gaussian and Laplacian pyramids, which, unlike previous work, does not rely on any explicit measures to distinguish texture from structure, but can effectively deal with previously challenging large-scale and high-contrast textures. Top: input images with diverse types of textures. Bottom: texture filtered results produced by our method.
We present a simple but effective technique to smooth out textures while preserving the prominent image structures. Our method is built upon a key observation—the coarsest level in a Gaussian pyramid often naturally eliminates textures and summarizes the main image structures. This inspires our central idea for texture filtering, which is to progressively upsample the very low-resolution coarsest Gaussian pyramid level to a full-resolution texture smoothing result with well-preserved structures, under the guidance of each fine-scale Gaussian pyramid level and its associated Laplacian pyramid level. We show that our approach is effective to separate structure from texture of different scales, local contrasts, and forms, without degrading structures or introducing visual artifacts. We also demonstrate the applicability of our method on various applications including detail enhancement, image abstraction, HDR tone mapping, inverse halftoning, and LDR image enhancement.

Overview of our approach. Given an input image \(I\), we first build its Gaussian and Laplacian pyramids {\(G_ℓ\) } and {\(L_ℓ\) }. Next, we upsample the coarsest Gaussian pyramid level \(G_N\) (\(G_N = R_N\)) to an intermediate texture smoothing image \(R_{N-1}\) at the previous finer scale. This is achieved by a pyramid-guided structure-aware upsampling (PSU) taking \(G_{N-1}\) and \(L_{N-1}\) as guidance. The resulting \(R_{N-1}\) is then subjected to the same upsampling process guided by \(G_{N-2}\) and \(L_{N-2}\). The above upsampling cycle is repeated multiple times until a full-resolution texture smoothing image \(R_0\) is finally obtained. \(JBF\) refers to joint bilateral filtering, and the symbol \(↑\) in \(JBF^↑\) indicates increase in spatial resolution of the output.

Results produced by our method. From top to bottom are input images and our texture smoothing results.
    title   = {Pyramid Texture Filtering},
    author  = {Zhang, Qing and Jiang, Hao and Nie, Yongwei and Zheng, Wei-Shi},
    journal = {ACM Transactions on Graphics (Proceedings of ACM SIGGRAPH 2023)},
    year    = {2023},
    volume  = {},
    number  = {},
    pages   = {}

Related Work
Hojin Cho, Hyunjoon Lee, Henry Kang, and Seungyong Lee. Bilateral texture filtering. ACM Transactions on Graphics 2014.

Zeev Farbman, Raanan Fattal, Dani Lischinski, and Richard Szeliski. Edgepreserving decompositions for multi-scale tone and detail manipulation. ACM Transactions on Graphics 2014.

Qi Zhang, Xiaoyong Shen, Li Xu, and Jiaya Jia. Rolling guidance filter. ECCV 2014.

Li Xu, Qiong Yan, Yang Xia, and Jiaya Jia. Structure extraction from texture via relative total variation. ACM Transactions on Graphics 2012.

Sylvain Paris, Samuel W Hasinoff, and Jan Kautz. Local laplacian filters: edge-aware image processing with a laplacian pyramid. ACM Transactions on Graphics 2011.