Structure-preserving texture smoothing with scale-aware intensity aggregation structure measurement

Abstract

Texture smoothing aims to smooth out textures in images while preserving prominent structure. However, facing complex images with multi-scale coexistence of texture and structure, the existing methods fail to distinguish small-scale structure from large-scale texture, which leads to undesired texture filtering. To this end, this paper proposes a novel scale-aware method. First, according to the texture and structure characteristics of one-dimensional signals, we propose a new texture metric, called intensity aggregation structure measurement (IASM), which has good performance in recognizing texture and structure. Second, we propose a structure-first-aware relative total variation, which can recognize important structural features with different sizes and shapes more finely, thereby estimating the calculation window of the new texture metric IASM for each pixel. Finally, the IASM with adaptive window cooperates with guided filtering to achieve smooth texture while preserving structure. The experimental results show that our method can protect high-quality structural features that are considered important visually, and at the same time, filter out large-scale textures well, which is better than existing state-of-the-art methods. Besides, our method is straightforward to implement and can be computationally efficient.