A Shape-Aware Structure-Preserving Texture Smoothing Algorithm

Abstract

In this paper, we introduce a shape-aware texture smoothing algorithm based on a flexible line shift scheme to separate prominent structures from textures or noises. Unlike previous texture smoothing algorithms, the line shift scheme can help to form shape-aware local area for each pixel during filtering. We integrate the line shift scheme with prior knowledge of prominent structures learned from a structured forests classifier into a framework of joint bilateral filtering. This new strategy can greatly improve the efficiency to remove textures/noise on pixels that are close to sharp corners or located in long narrow strip areas, while keep the prominent structures untouched. We demonstrate the performance of the proposed algorithm on a wide variety of images corrupted by textures and Gaussian noise.