Semantic-Aware Image Smoothing

Abstract

Structure-preserving image smoothing aims to extract semantically meaningful image structure from texture, which is one of the fundamental problems in computer vision and graphics. However, it is still not clear how to define this concept. On the other hand, semantic image labeling has achieved significant progress recently and has been widely used in many computer vision tasks. In this paper, we present an interesting observation, i.e. high-level semantic image labeling information can provide a meaningful structure prior naturally. Based on this observation, we propose a simple and yet effective method, which we term semantic smoothing, by exploiting the semantic information to accomplish semantically structure-preserving image smoothing. We show that our approach outperforms the state-of-the-art approaches in texture removal by considering the semantic information for structure preservation. Also, we apply our approach to three applications: detail enhancement, edge detection, and image segmentation, and we demonstrate the effectiveness of our semantic smoothing method on these problems.