局部尺度自适应纹理滤波

Local Scale-Adaptive Texture Filtering

摘要: 针对传统方法在图像纹理滤波过程中难以区分结构边缘和纹理，易将高频纹理误判为结构边缘，导致显著纹理无法被平滑的问题，提出了一种局部尺度自适应纹理滤波算法。首先通过多方向侧窗滤波技术确定最优的结构边缘特征。随后，基于边缘强度计算局部尺度信息，通过归一化处理得到每个像素点的自适应核尺度。进而利用自适应高斯滤波，实现初步的纹理平滑，得到指导图像；接着利用联合双边滤波进行精细化的结构保持处理，增强纹理滤波结果。实验结果表明，本文提出的方法在有效抑制多种纹理模式的同时，能够更好地保留图像的主要结构。

Abstract: To address the issue where traditional methods struggle to distinguish structural edges from textures during image texture filtering, often misclassifying high-frequency textures as structural edges and failing to smooth prominent textures, this paper proposes a locally scale-adaptive texture filtering algorithm. First, multi-directional side-window filtering technology is employed to identify optimal structural edge features. Subsequently, local scale information is computed based on edge intensity, and normalized processing yields an adaptive kernel scale for each pixel. Adaptive Gaussian filtering is then applied to achieve preliminary texture smoothing, yielding a guide image. Subsequently, combined bilateral filtering performs refined structural preservation processing to enhance the texture filtering results. Experimental results demonstrate that the proposed method effectively suppresses various texture patterns while better preserving the primary structural features of the image.